However, I will say that the expected value of the p2pool payout method (if set to zero fees) is higher than the expected value of the eligius payout method for finite time.
Can you derive this for me? I've just spent a bit of the afternoon on it and I can't see it.
First, let's ignore orphans, they affect both more or less equally, though how exactly they affect eligius is complex. It doesn't help refute the argument though.
The expected value of every share on p2pool (PPLNS) is exactly B/D (for some suitable scaling of B and D). This can be derived, and is likely is derived in one of Meni's posts.
The expected value of a share on eligius is (the product of an equivalent B/D and) the sum of an infinite series adding the probabilities that it is paid on this round plus the probability that it is paid on each future round. This series never goes to zero, so the terms beyond any finite time horizon always have a positive sum.
Not formal, but good enough?