Indeed, but that's what I'm having a hard time understanding. To my mind, if the compiled application knows to throw 64-bits at the processor at a time instead of 32, how could that be anything other than a faster way to process instructions and data? Admittedly, given the notion of bottlenecks getting 'filled to capacity' probably not twice the processing speed, but certainly faster than the 32-bit version.
It would require an application written to work in a 64-bit environment and for Tiger to support it, but I don't see how this would not improve performance on a G5 system as and when the software exists.
Clearly I'm missing something!