There is actually a similar question that is asked in the interview:

> Is there anything in the kernel which is not optimal, but would require a complete rewrite to address properly? In other words, the kernel is 30 years old and knowledge, languages and hardware have changed a lot in these 30 years: if you rewrote it from scratch now, what would you change?

And Linus' answer was very interesting for me - and after reading it, I would answer your question (tongue in cheek): "Never, because it seems that Linux developers were quite successful in starting all over again for quite some parts of the Linux kernel."

I guess it's not really an answer to your question, but kind-of ;)