This is a very good point. Certainly I’ve been critical of Apple’s lack of interest in improving the scheduler performance of their kernel (not sure how much is even possible given the Mach architecture that is simply not ideal for realtime applications), but as with so many other performace optimization issues, ultimately they’ll become moot as time marches on and the performance margins vanish.
There’s an extreme example of this in CPUs, as Intel’s chips used to be at a disadvantage to RISC CPUs because of the more complex instruction set that requires more space on the chip just for processing the instructions… well, can’t say it hurt them that in the long run! And no one even notices the extra baggage on the chip anymore, except in phone/tablet hardware, although even that’s going away.
Well, this is slightly different of course. But while the performance gap still exists (my recent benchmark test was around 25% in favor of Windows, so the 20-30% range seems very accurate), there’s likely going to be a point where it’s not going to be terribly important anymore. To make a better sense of those prospects though, it would be interesting to see how the performance gap scales across old and new systems, but alas I don’t have the time or means to test that extensively.