I'd say build first, optimize later. Once you have a working prototype, check to see if the explosion looks like it's gotten out of hand. My guess is that limiting the number of selectors won't affect the compressed download size that much. If it does, then we can always choose the best one (or two) methods for selectors.