As I said, one way or another, one chain (fork) will survive. Users on the other chain may scream bloody murder, but arguing with math will get you nowhere.
Yes and if sufficiently mixed, you can't try appease those who want the bad fork, because you can't extract their transactions from the bad fork and put into the good fork.
And this is the qualitative threat difference from block chains that don't mix transactions.
Except that all chains have mechanisms of mixes, maybe not on chain, but good luck untangling any block chain after any significant period of time, once people have traded through exchanges (many that are effectively totally anonymous), used coin mixers, used coins to rent rigs and mine new coins, etc. You can probably do it for a small number of blocks, just as a fork of around 40 blocks caused no lasting trouble for Monero last month. But after hours or days, any chain is equally intractable to undo.
Furthermore I'm not convinced even if it could be done, that it would be helpful to users. Fungibility might very well be more valuable than the ability to pick winners and losers after an incident.
True that any coin mixing (i.e. not IP obfuscation mixing) is qualitatively equivalent, but coin mixing on chain could be explicitly denied by only allowing one input to a transaction (and dedicated means to merge balances would be needed) or the user could selectively agree not to use on chain mixing.
Non-decentralized mixers can mix coins with out multiple inputs per transaction, but these can't be trusted thus in my mind they are not anonymity any way.
Some transactions are unwound. Unwinding less transactions by being able to segregate transactions in the attacked fork and add to the non-attacked fork those which are not downstream from a double-spend or stolen coinbase afaics doesn't decrease fungibility? Rather it aids a potential political consensus to choose the non-attacked fork, i.e. afaics it adds fungibility.