The fourth scenario is more interesting because we've admitted the distribution of p is unknown to us. You note that if p is uniform, all intelligent lifeforms should conclude they live in a universe where p is high, but I'm not certain of that. It seems to me that this logic only works if you conclude there is _one_ universe for each possible value of p. But if, for each possible value of p, the multiverse contains universes for each possible outcome of a Bernoulli trial with probability p, then we can't conclude anything at all. To see this, notice that the number of Lucky universes (where p is high) is uncountably infinite, just as the number of Unlucky-Lucky universes (where p is low but a miracle has occurred) is uncountably infinite.
I haven't thought this next part through as deeply, but I believe this problem compounds given the unknown distribution of p. This is because subjectivist models of probability (e.g. Bayes' Theorem) are heavily influenced by the distribution of the prior when we have only observed one trial. And in cases like this one, where the only trials that _could_ be observed are successes, I believe it's all but entirely driven by the distribution of the prior. (You could imagine a hyperprior - a distribution of distributions of p, but I don't think we have any good reason for deciding what that would be either).
Anyway, thanks for this. I love geeking out about epistemology.