What's wrong with relying on A/B testing to inform design? In my opinion, except for the whole Google+ unifying process going on most Google products are highly usable.
It can help you find local maxima, but not the global maximum you are (or should be) looking for.