This is something I don’t see talked about much, but the asymptotic state of Bitcoin when mining fees are zero is interesting to think about. It’s not obvious to me that the economic incentives against running a 51% attack will exist in the “terminal” stage of Bitcoin in the way that they do in the “inflationary” (positive mining reward) stage.
Especially if, as you point out, layer 2 networks become a more desirable place to do transactions.
Basically, it seems like the terminal state of this is a gigantic freeloader problem where everyone wants the network to be secure but nobody is incentivized to do it.
It'll be in the interest of exchanges and payment processors to prevent double-spend at least, and they have an incentive for the blockchain to be trustworthy vs. running their own attacks and could adjust their own fees to maintain >51% control of miners across the lot of them.
There's also a fallback recovery mechanism; for any attacker willing to invest $X in a 51% attack the remaining users only have to invest an additional $X*.04 to regain 51% control and make a hard fork that most miners would switch to. My guess is that faced with evidence of a double-spend most miners would switch to the hard fork to avoid losing the block rewards the attacker's chain stole from them. This relies on rapid detection of the attack and availability of a hard-fork client.