Honestly, I don't think so :) in 32bit, each program can access 2GB of memory. most program use less than 2GB of memory I dont see why larger registers by itself should be faster...
unless, and this is just thoughts, the program allocates lots of small memory pieces and deallocates them again in a pattern that creates a lot of memory fragmentation. Hmm but I don't think this would actually be 'slower' with less memory, you would just hit the wall faster (according to avaliable memory size)...
