My mistake. I was referring to x86-64. I knew Intel tried to rename it from AMD64 to something closer to IA-32 which was their own name for 32-bit x86. Apparently they want us all to call it Intel 64. Anyway as I pointed out above the extra registers are nice, but real world performance is not massively effected by them, especially in a heavily IO dependent app like the Finder.
Sure, the benefit of those extra registers might not be earth-shattering, but it's still a good thing to have. And I think that those extra registers are enough to offset any performance-hit 64bitness might introduce when compared to 32bit. In PPC (for example) 64bit might be slower in some cases due to the fact that architecture remained pretty much unchanged. But in x86 we get more registers (something x86 desperately needs), and that should negate the potential performance-hit 64bit introduces.
IIRC, I remember back when AMD released the Opteron there were benchmarks that showed that those extra registers give 3-10% performance-boost when compared to 32bit x86, even in cases where the application doesn't really benefit from 64bit as such.