Yes. Except nobody does that anymore, because it's a waste of time with processors as fast as they are today, and compilers being as good as they are.
I guess you are one of those nobody's. This is an unfortunate and lazy attitude to take - you should always look at the opportunities to make core and important routines faster - especially in game programming. As fast as our processors are, some of our bloated software should run a LOT faster, if some extra care was taken.
What gives you this idea that nobody does this anymore? Take a look at Intel's math libraries - these are hand written assembly routines, to better utilize the vectorization and super-scalar capabilities of their processors. I'm sure Apple have similar libraries to use the Altivec core.
Indeed, compilers are very good today, however they are far from perfect. I'm very interested in the coming revisions of gcc, which will implement vectorization. None of the main stream commercial compilers use the extended registers of the x86 platform to any great advantage - they stick them in for filling or clearing memory from time to time, but that's about it. Intel is better than Microsoft and does some simple vectorization, but there are still many missed opportunities. One must either use compiler intrinsics or drop down to assembly to fully take advantage of this architecture.
This doesn't apply to every day programming, or even a large amount of game programming, however multimedia utilises it significantly.
My point stands, ASM level optimizations (such as Intel's math library) can be reused across platforms.
What will be cool is the dual-boot possibility of an Intel-based Mac. I can have one Mac laptop, that can dual boot OSX or a Windows based OS.
VMWare will no doubt release a version of it's software for OSX now, so you can run x86 based OS's "natively" in OSX. Natively=x86 instructions will no longer need to be emulated, like VirtualPC. Very cool stuff.
Cheers,
Stu