Have you considered compiling the Windows port of
MMCache with Intel C++ Compiler 7.1 in order to take
advantage of SSE2, SSE and MMX instructions
In my tests ICL performs about 25% better than
VC++ 7.1, even in my Athlon XP CPU! The advantages
comparing it with VC++ 6.0 should be greater.
It should be as easy as recompiling with ICL adding
the switches /G6 /O3 /QaxWKMi or even better
distribute different executables for MMX
(/G6 /O3 /QxMi), SSE+MMX (/G6 /O3 / /QxKMi) and
SSE2+SSE+MMX (/G7 /O3 / /QxWKMi).