I still prefer the version Linus recommended. It's not for speed - I think it's more understandable. It's more simple (in the way Richard Hickey says, one strand rather than many) and there is less of it. Proving it correct would be easier, not that I actually would.
When people say "clever" to me, I think of code that is being tricky about something. (My canonical example of clever code is Duff's Device.) Linus' example doesn't read as tricky to me, it reads as very straightforward.