Hawking's path integral methods seem to rely on the assumption that superpositions of different metrics are meaningful. (If I'm wrong about this, let me know). But are they? Aren't these superpositions destroyed by decoherence. And aren't they also in contradiction with Penrose's claim that there is no well-defined time evolution operator on a superposition of two spacetimes, and that such superpositions should therefore be highly unstable?