If we consider *really small* fees, then for an extra included transaction, that extra delay on the network will mean an extra probability of the block being orphaned, putting in jeopardy the whole income. This even happens with small blocks.
Yes, that's probably the way Peter R's model was intended to work (though it not even works out in theory).
Furthermore, as long as you have a block reward that is significantly higher than the total fees per block, even a small increase of the orpahing risk might deter a miner from including every transaction.
Yes, this will simply result in cutting off the very lowest fees of the fee distribution, which will remain for ever in the mem pool.
I don't think that this has much to do with "optimal size" ; it only means that one doesn't include the cheapest transactions below a given fee threshold, because their extra transmission time penalises the whole income while not contributing enough to it.
I don't think that this has much to do with "optimal size" ; it only means that one doesn't include the cheapest transactions below a given fee threshold, because their extra transmission time penalises the whole income while not contributing enough to it.
I think that's the whole point of it. If the model worked as intended, the market would agree on a minimal fee for a transaction to get included in a block. It's not about finding the "optimal size" but the optimal price.
That said, a market doesn't need to come to "equilibrium". An erratically chaotic market dynamics can be fun too 
It would be indeed fun to see the majority of nodes switching to Bitcoin Unlimited...