I think the theorem implicitly assumes logical omniscience, and using heuristics instead of doing explicit expected utility calculations should make sense in at least some types of situations for us. The question is whether it makes sense in this one.

I think this is actually an interesting question. Is there an argument showing that we can do better than prase's heuristic of rejecting all Pascal-like wagers, given human limitations?