[marcosdumay]

That's because kernels perform differently from user software (they have more optimizations available), and a lot of big code needs that extra performance.

That's a fact that still didn't change, but user level code is getting more powerful (mainly for virtual machines), and computers are still getting faster. So, it's still too early to declare the race finished.

Anyway, none of that has any relevance to how one should organize user level code.