I'm also don't know how you could actually verify that a node had indeed previously seen a transaction it was not broadcasting. Each node would have to store and share a list of every transaction it was broadcasting to every node, but even then, there would be no way to verify that a node actually received what was being broadcast to it.
Of course a node could not verify which transactions the other node
saw. What each node could verify, in contrast, are the transactions which were
signed. That whole mechanism only makes sense combined with the "challenging blocks" mechanism I described earlier.
I'll probably wrap my head a bit around that mechanism. I suspect there could be a way, based on such a mechanism and maybe a slight PoS component, to force miners to include only the transactions with most fees, or at least provide very strong incentives to do so. You're right that this would be much more intensive computationally, but maybe it is worth it thinking about it anyway.
I also don't think it's particularly necessary. One or two nodes participating in censorship is irrelevant - transactions will spread around them without issue. Even if we pick some ridiculous number like 50% of nodes which start censoring transactions, [...]
The mechanism I was referring to is directed more to miners than to normal nodes. Miners unfortunately could start to censor massively, like it seems to have occurred in Ethereum. Thus the idea to punish them for misbehaving (censoring, in this case).
Anyway, this discussion is mainly a theoretic one. I believe that in general Bitcoin's incentive mechanisms work well. But it's not wrong, I think, to try to imagine possible bigger improvements for the long term.