It depends what your using them for, but anything with smooth in their name is better. These come from Frank Dimiao. Not sure if they are published anywhere, but the stats he used to regenerate them are of a much much larger set of PDBs and he used techniques to smooth out the surface (Similar to what Roland/Max used for the 2010 Rotamer library). If you have a choice, I highly recommend using these. Most centroid-based protocols have been optimized through the old centroid scores, so if using these in those, you may want to run some sort of benchmark first.
.
