One of the reasons to pick a longer window of, say, a couple difficulty periods would be that you can make participation in the pool hedge you against hashrate changes.
You're absolutely spot on to think about the impact of pooling w.r.t. variance when fees > subsidy. That's not really in the analysis I had in the (old) post, but when the block revenues swing, dcfmp over longer periods can really smooth out the revenues for miners in a great way. This can also help with the "mind the gap" problem when there isn't a backlog of transactions, since producing an empty block still has some value (in order to incentivize mining transaction at all and not cheating, we need to reward txn inclusion as I think you're trying to point out.
Sadly, I've read the rest of your email a couple times and I don't really get what you're proposing at all. It jumps right into "things you could compute". Can you maybe try stating the goals of your payout function, and then demonstrate how what you're proposing meets that? E.g., we want to pay more to miners that do x?