ST_Distance_Sphere is implemented only for points.
I suggest you to use ST_Distance instead, which provides the minimum distance between two bi-dimensional geometries of any type.
I've performed a similar test with two tables (point vs polygon), using this code:
SELECT a.gid point_gid, b.gid polygon_gid, (SELECT ST_Distance(a.geom, b.geom)) distance
FROM harta_gal_etichete a, harta_gal_parcele b
WHERE (SELECT ST_Distance(a.geom, b.geom) < 50)
ORDER BY a.gid;
If you have really big tables (spatial indexed), you'll have to wait a few minutes to obtain a result like this:

EDIT:
Thanks to the comments I was able to reproject my code and to achieve a significant reducing of the (same data set) processing time (from 269 seconds to 10):
-- Reproject the point table geometry
CREATE TABLE public.harta_gal_etichete_4296 AS
SELECT ST_Transform(geom, 4269) AS geom
FROM harta_gal_etichete;
-- Reproject the polygon table geometry
CREATE TABLE public.harta_gal_parcele_4296 AS
SELECT ST_Transform(geom, 4269) AS geom
FROM harta_gal_parcele;
-- Compute the distances between the two geometries
SELECT a.gid point_gid, b.gid polygon_gid, (SELECT ST_Distance(a.geom, b.geom)) distance
FROM public.harta_gal_etichete_4296 a, public.harta_gal_parcele_4296 b
WHERE ST_DWithin(a.geom, b.geom, 50)
ORDER BY a, b

I choosed this solution because using ST_Transform for 205496 rows last forever.
Full credit goes to "How to reproject all geometries in a PostGIS table", Taber and Jakub Kania