Not yet.

We have even had reports that native 64bit is actually slower for this kind of computation. Stick with the emulation for now.

http://www.theinquirer.net/default.aspx?article=34576

CPUs began implementing 64-bit paths back with the 80486. Today the question is whether or not the cache works with 128 bits or 256 bits. Meanwhile, there are lots of 64 bit and even 128 bit registers for SSEx instructions. The final switch to explicit 64-bit ops and registers (and doubling the number of registers) has speeded up programs, but much less than you might imagine unless you consider all the evolutionary developments within the CPU this last decade.

Last year I read a white paper comparing a new 64-bit math library using carefully optimized hand-coding for AMD64 which compared it with the earlier 32-bit library. While it varied from function to function, the general speedup was on the order of 10% to 15%. Since compiler optimization tends to develop at a disappointingly slow pace, I suspect (my personal opinion only) that the actual speedup with current compilers will be 10% or less (probably less). The important exception is with large databases running in greater than 4 GB of RAM. But we are not downloading large databases with current projects.

I am still waiting for benchmarks using the new Intel Core 2 CPUs. I have read that they optimized their macro-ops fusion for 32-bit instructions, which might imply that they show less of an advantage running in 64-bit mode against AMD64 than they do in 32-bit mode against AMD32.

The future belongs to large memories running 64-bit programs that can access all the memory, but for right now small programs that fit within 4GB ( ) probably run almost as well in 32-bit mode as in 64-bit mode.

Just my opinion,
Lawrence