Probably you've noticed that primarities are $\mathbf{X}$, $\mathbf{Y}$, $\mathbf{Z}$, not $\mathbf{R}$, $\mathbf{G}$, $\mathbf{B}$ (which are corresponding to the color values $R$,$G$,$B$). This is the aftermath of original work conducted by Wright and Guild yielded the Color Matching functions: $r(\lambda)$, $g(\lambda)$, and $b(\lambda)$ having a negative value of the red primarity around $522 \;\texttt{nm}$ (figure below). Everyone will agree that it is non-intuitive and somewhat confusing.
Why values of $r(\lambda)$ are negative? Well after tristimulus color matching experiments, it turned out that not every color can be created with the preset primarities ($700\; \mathtt{nm}$, $564.1\; \mathtt{nm}$, and $435.8\; \mathtt{nm}$). So for example, if you wanted to create the blue-green color of wavelength $\lambda = 490 \; \mathtt{nm}$ it cannot be done only by adding blue and green primarity alone. You also need to subtract the red one. Mathematically speaking, the color equation:
$$\mathbf{C}=r\mathbf{R} + g\mathbf{G} + b\mathbf{B} $$
Must by modified by adding the outer mixture $-r_{out}\mathbf{R}$ to both side of the above equation:
$$\mathbf{C} + r_{out}\mathbf{R} = g\mathbf{G} + b\mathbf{B} $$
That is why CIE introduced new set of so-called virtual primarities: $\mathbf{X}$, $\mathbf{Y}$, $\mathbf{Z}$. In order to obtain only positive or null values of the Color Matching Functions. That's how $2^{\circ}$ observer Standard Color Matching Functions (also called CIE 1931) were created. By doing so they ensured that (amongst other assumptions):
- New SCMF's $x(\lambda)$, $y(\lambda)$, $z(\lambda)$ have values $\ge 0$ (not like $r(\lambda)$).
- Are related to previous CMF's ($x$,$y$,$z$) by a appropriate transform.
- $\mathbf{X}$, $\mathbf{Y}$, $\mathbf{Z}$ must be independent (kind of orthogonal).
- Only color value $Y$ is proportional the the lightness of a colour.
Newly obtained color quantities $X$,$Y$,$Z$ are called Standard Color Values. The relation between CMF's and SCMF's is given by a following linear transform:
$$
\left( \begin{array}{c}
x(\lambda) \\
y(\lambda) \\
z(\lambda) \\
\end{array} \right)
=
\left( \begin{array}{ccc}
2.7689 & 1.7517 & 1.1302 \\
1 & 4.5907 & 0.0601 \\
0 & 0.0565 & 5.5943
\end{array} \right)
\cdot
\left( \begin{array}{c}
r(\lambda) \\
g(\lambda) \\
b(\lambda) \\
\end{array} \right)
$$
After over a three decades people from CIE had realised that $2^{\circ}$ observation angle might not be enough for adequate color assessment. That's why in 1946 they introduced the $10^{\circ}$ observer, also known as CIE 1964 observer. Latter is recommended for visual angles above $4^{\circ}$, whereas CIE 1931 should be used for any angle below that. Of course there is no sudden change in perception of colour at $4^{\circ}$, this number is just arbitrary. Comparison of these two standards is depicted of a following plot, on which you can see the 'bump' mentioned by you for the $x(\lambda)$ standard color value:
