> In the ES paper, and in your responses, the assumption is a colluding group of miners. I find myself agreeing with you there; it looks like the colluders have more incentive to break ranks. But suppose a single individual controlled 33%. Couldn't they use this same strategy to benefit more than we previously thought they could?

Yes, I think they could. I'm currently trying to model what happens in mining pools where the pool master attempts to keep the pool's state a secret from the pool members to deter the kind of hopping behavior necessary for fair weather mining.