I've love to able to either measure or use all this lovely information in my running Linux desktop GUI app. But I happen to know that would probably not be worth my time except some rather specific cases. Plus the direction of PC architecture seems to be towards more and more levels of cache and that implies more and more work if you actually take these into account.

Just finding the time that library calls take is quite a pain in Linux.

Given all this, I'm inclined to think that I'd rather architect cache-oblivious data structures than root-around endless when background threads and other stuff suddenly get slow...