More general-purpose registers in such processors let you optimize work with local variables in functions which do not need to be saved into main memory anymore. But in general you may expect a 2-20% performance gain from mere recompilation of a program - this is explained by architectural changes in 64-bit processors. The difference in performance between 32-bit and 64-bit versions of applications greatly depends upon their types and data types they are processing. Why do 64-bit applications work faster than 32-bit ones?
