Hello charitylover,
This is a perennial question. We would need to debug Rosetta before maintaining different optimized versions made much sense. But then - - - last year I read a paper in which some developers had carefully hand-optimized a math library for 64-bit code. They said that the routines usually ran 10% - 15% faster. Since it takes time to design a good optimizing compiler, and the results are usually not as good as manual optimization, I would assume that the maximum speedup would be in the range of 10%. Quite possibly less. Eliminating the current crop of bugs would probably pay better dividends right now. But eventually, with improved compilers that people have spent a few years on and with more members that can use such code, it might prove a very useful step.

Lawrence