Lots of companies are working on virtualization. E.g. AMD and Intel are adding extensions to their newest chips to allow multiple OSes to run "simultaneously". There's overhead with such approaches, obviously, and even more so if the program assumes a different instruction set.

It also introduces the "Win-OS/2" problem again. If the virtualization is good enough, there will be little incentive for ISVs to write to the native CPU. Then Windows x86 wins by default. If it's not good enough, then the migration to the new CPU or OS is too painful for customers with existing vital software. Then Windows x86 wins by default.

It's tough for translation/virtualization to win unless the underlying CPU is ~ 10 times faster than the incumbent. The Cell might have a chance, but it's hard to see that virtualization of existing MacOS X software on x86 variants being a compelling solution.

My $0.02.

Cheers,
Scott.