In terms of product development, perhaps Google would be better off implementing a Y-Combinator based model where 2 to 5 people work on products by themselves -- away from Google's offices. They'd do this anywhere they'd want -- living together in a house/apartment, working from coffee shops, etc. They'd have access to Google's resources if/when they'd need them (kind of like how YC now has significant resources to help out its companies -- e.g. TechCrunch for initial distribution), but would not be required to use them.
The ultimate metric for a web property is getting repeat users at scale, so that's the #1 goal.
But as you say, Google doesn't really suffer from this problem at least initially (they can always get a substantial amount of immediate users for any new product).
The problem is, are they products people want? Are they products that'll generate repeat users at scale?
I wonder if the people building products at Google are building things for themselves, or for what they think people want.
The best products are made with one user in mind, and that user is the maker.