I agree that estimation questions (not "brain teasers" as coming up with the clever solution) are good. Developers should be able to think in orders of magnitude.