For the kind of developers I like to hire, speed, performance, and quality all just happen. It doesn't take any longer to write quality high speed code than it does to write garbage that just happens to get the job done.
I wouldn't set time limits on such tests... I'd just tell them to pretend they're writing code that will go into production and to give it their absolute best shot.
As for the FizzBuzz test... it's a simple and well published solution so if I ever used it as part of a interview, I'd have to pork chop anyone who uses a loop or recursion to solve the problem. I'm not looking for someone who can't even count in a setbased fashion.
Usually though... it doesn't take a test to be able to tell.
--Jeff Moden
Change is inevitable... Change for the better is not.