My copy of Player’s Handbook has that statement on page 171.
Bonus Types: Many spells give their subjects bonuses to ability scores, Armor Class, attacks, and other attributes. Usually, a bonus has a type that indicates how the spell grants the bonus. For example, mage armor grants an armor bonus to AC, indicating that the spell creates a tangible barrier around you. Shield of faith, on the other hand, grants a deflection bonus to AC, which makes attacks veer off. (Bonus types are covered in detail in the Dungeon Master’s Guide.) The important aspect of bonus types is that two bonuses of the same type don’t generally stack. With the exception of dodge bonuses, most circumstance bonuses, and racial bonuses, only the better bonus works (see Combining Magical Effects, below). The same principle applies to penalties—a character taking two or more penalties of the same type applies only the worst one.
(emphasis mine)
Note that both the SRD and Player’s Handbook have this information, bizarrely, in the magic overview section. Arguably, this placement means that this stacking applies only to racial bonuses given as part of spell effects. Unlike d20srd.org, the Player’s Handbook as well as “Basics.rtf” from Wizards’ own publication of the SRD does not include information about modifier types or stacking, at least not until you get to the above-quoted section on spell bonus types, which suggests that, despite its placement in the magic overview, it is the general rule. Moreover, on page 172, we have the “bonus types” section referred to as being the general rule beyond spellcasting:
More generally, two bonuses of the same type don’t stack even if they come from different spells (or from effects other than spells, see Bonus Types, above).
So yeah, weird placement, really confusing and easily missed, but it appears to be official that racial bonuses stack with each other.
On the other hand, page 21 of Dungeon Master’s Guide—which is the section on modifier types and stacking found in the basics section of d20srd.com—lists “racial” as a bonus type with no mention of its self-stacking. The default—established in that very section, though also mentioned in the Player’s Handbook Bonus Types section on page 171—is therefore that they would not stack.
Primacy: Dungeon Master’s Guide?
The official errata documents for D&D 3.5e provide guidance on what to do when official rules sources contradict each other like this. The following paragraph is found in the introduction to both the Player’s Handbook errata and the Dungeon Master’s Guide errata:
Another example of primary vs. secondary sources involves book and topic precedence. The Player’s Handbook, for example, gives all the rules for playing the game, for PC races, and the base class descriptions. If you find something
on one of those topics from the Dungeon Master’s Guide or the Monster Manual that disagrees with the Player’s Handbook, you should assume the Player’s Handbook is the primary source. The Dungeon Master’s Guide is the primary source for topics such as magic item descriptions, special material construction rules, and so on. The Monster Manual is the primary source for monster descriptions, templates, and supernatural, extraordinary, and spell-like abilities.
The primary source on “playing the game,” which something as basic as bonus types would fall doubtlessly fall under, is the Player’s Handbook. Bonus types are certainly not magic items or special materials, so the explicitly-listed examples of things primarily handled by Dungeon Master’s Guide don’t apply. But that list then does say “and so on,” and the actual Dungeon Master’s Guide modifier and bonus types discussion seems far more “primary”—it’s right at the front of the book, in its own section, as opposed to buried in the magic overview on page 171.
More importantly, the Player’s Handbook page 171 section explicitly says “(Bonus types are covered in detail in the Dungeon Master’s Guide.)” What does that mean? It could mean that the Player’s Handbook is expressly indicating that the Dungeon Master’s Guide rules are primary. As the primary source on something, can the Player’s Handbook delegate that primacy to another source? This is getting absurd, and it’s not really clear that this is what’s going on or that it can do that, but I’m not really going to argue against it—we’re already pretty far down the rabbit hole, it’s not really legit to arbitrarily decide that here is where we are going to draw the line.
Specificity: Player’s Handbook?
There is also a question of specific vs. general. It’s not explicitly covered by the errata introduction (to my surprise; I thought it was), but 3.5e is an “exception-based” ruleset, meaning the basic rules apply general facts (e.g. typed bonuses don’t stack with bonuses of the same type) and then more specific features provide exceptions (e.g. dodge bonuses to AC do stack with each other). So which of these is more specific? And does being more specific restrict it to only a particular realm?
For example, if a spell had said “this racial bonus stacks with other racial bonuses,” that would be more specific—no one would question that it overrides the default bonus-stacking rules—but it’s also more narrow—it only applies to the racial bonus from that spell. The question here is, how do the sources of rules for racial bonus stacking compare with regards to specificity?
The placement of the Player’s Handbook rule on page 171, in the magic overview, could arguably make it a more specific case—so it would override the Dungeon Master’s Guide rule and allow racial bonuses to stack—but again a more narrow one—not applying to racial bonuses from, say, races or templates, which would still not stack. However, the statement on page 172 indicates—explicitly—that the Bonus Types section on page 171 is general, and does not apply only to magic. So officially, it’s not meant as a specific exception—it’s supposed to be the general rule. That doesn’t help it overrule the Dungeon Master’ Guide rule, but it also means it’s not restricted solely to magic.
So does that answer the question of specificity and send us back to considering primacy? Actually, no: for two reasons. First of all, it’s rather dubious for page 172 to “generalize” the rule on page 171 like that. Usually, more-general rules should be found in more-general sections and use text that is, well, general. Half of the Player’s Handbook page 171 Bonus Types section is talking about spells. It’s not clear that page 172 is allowed to generalize that. But they wrote it that way, and the relevant portion of the Bonus Types section is written generally, not specifically about spells, so I think it should apply.
The other concern is somewhat more important, though: the Player’s Handbook description includes a specific exception for racial bonuses. The Dungeon Master’s Guide does not say specifically that racial bonuses do not stack—it says generally that bonuses don’t stack, and does not provide a specific exception for racial bonuses. But any claim that “the Dungeon Master’s Guide rule is that racial bonuses don’t stack,” is relying on the general rule. Arguably, the Player’s Handbook specific exception for racial bonuses overrides the general Dungeon Master’s Guide rule, not a similarly-specific rule about racial bonuses.
Conclusion
RAW, we have a mess:
The Dungeon Master’s Guide section is probably the primary source on bonus types—the errata indicates it should be the Player’s Handbook section but that section redirects us to the Dungeon Master’s Guide section. There are some counterarguments that could be made about that, but since the Player’s Handbook rule is so buried and the Dungeon Master’s Guide is right up front, I’m not inclined to make them.
However, specificity favors the Player’s Handbook section. This isn’t the strongest case ever—the Dungeon Master’s Guide Racial Bonuses section is, probably, the primary source on racial bonuses, and its lack of rule should probably override the Player’s Handbook inclusion of racial bonuses in a single sentence. Probably; maybe. But strictly speaking, the Dungeon Master’s Guide position relies on falling back on the general rule, while the Player’s Handbook rule explicitly refers to racial bonuses as an exception to that very rule. And both page 171 and the errata tell us this is where we should look.
Ultimately, though, all of this analysis is kind of nonsense. If it came to some solid conclusion, that’d be one thing, but it doesn’t. The reality is, someone at Wizards of the Coast made a mistake—presumably, either racial bonuses once stacked and someone forgot to update the Player’s Handbook mention when that changed, or else racial bonuses once did not stack, and someone forgot to update the Dungeon Master’s Guide section when that changed. On the one hand, the Player’s Handbook rule is buried—easily missed. On the other hand, the 3e Player’s Handbook didn’t even have the Bonus Types section—the only mention of bonuses stacking or not stacking was in the combining magical effects section (that part that 3.5e has on page 172 and that refers back to 171), where stacking racial bonuses were not mentioned. So was the addition to the Player’s Handbook rule meant to be included in the Dungeon Master’s Guide rule, or did they add it and then decide against it and forget to remove it from there? We have absolutely no way of knowing.
For my own games, racial bonuses don’t come up that often. They usually come with pretty hefty costs, usually in the form of level adjustment. I can’t think of any particular need for non-stacking racial bonuses. I have no real problem with them stacking, so I allow it. If someone was using a template and race combination that each provided a racial bonus to the same thing, it would seem feel-bad to make them lose one of them, considering the costs associated with templates and racial choices. But then, I don’t generally allow most templates anyway—level adjustment is not usually available to my players at all.