About the idea of “punisheable Eltoo” in the context of multi-party setting, this has been already explored a bit in the past by Lloyd Fournier and also myself. It’s not exactly the same concept than you’re proposing if I’m understanding what you’re trying to achieve, though here few links if this can be of interest for your research:
There is also this post too on the challenges of multi-party construction, whatever the security model, in a pure trustless design (no coordinator, no quorum) though interesting for your current research.
For the question about “how many parties we’re really shooting for in a multi-party channel”. For the CoinPool paper, we went with 1000-sized pool to get 7.9 billion of human beings each with its own pool balance. See the assumptions we were making on the bitcoin blockchain (no consensus change in validation ressources — all other things equals apart covenants) in this post.
From a brief read of your research, the hard limitation I can see is not actually the one chance to update the channel state on-chain (though how do you assign blame correctly among N non-trusted counterparties in an efficient fashion is an interesting problem ?), but rather the exponential growth in the vector commitment with the number of counterparties participating in the off-chain construction.