Krakatos wrote:
Gains of 30% over the 32-bit kernel were found in these tests. And that's a huge difference.
Look again. The big differences were between Leopard & Snow Leopard, with 64-bit vs. 32-bit SL differences substantially smaller.
Also note that these tests were done on a Mac Pro with a quad-core "Nehalem" CPU, 12 GB of memory, & a 4 drive RAID. Don't expect much similarity between tests run on that very high performance setup & anything you would see with an iMac or especially a MacBook (Pro or otherwise), even ones with the fastest available Core 2 Duo CPU's & maxed out memory.
Even though all these Macs use 64 bit Intel CPU's, the Core 2 Duo & Nehalem architectures have substantial differences, among them being the latter's "Turbo Boost" technology (which can dynamically boost the clock rate of a core if the temperature does not climb too high), fully buffered ECC DIMMS (which may require oversized heatsinks for reliable operation, according to
Ars Technica), "QuickPath Interconnect" (a point-to-point interconnect that replaces the old front side bus technology), & many others.
Basically, all these posts that start with the premise that all it takes to get real benefits from the 64-bit kernel is a 64-bit CPU start with a flawed premise. It isn't just about wider data paths or addressing more memory. It is also about the details of how the CPU implements the 64 bit architecture (like internal bus speeds, cache sizes & levels, instruction set additions, & even more arcane stuff), how the system gets rid of heat, & much more.
It is a lot like saying that all V-8 engines are very much alike; therefore, they all are capable of developing 4000+ horsepower because some "Funny Car" class V-8 powered dragsters can, so there is no reason for car makers not to let us use more of that power latent in our V-8 powered family cars.
You can't use what isn't there.