But you are arguing that the attacker doesn't need a majority of the historic private keys (to overcome TaPoS), so therefor the number of attackers are unbounded. Thus there is no objectivity between unbounded forks. Thus of course the current stake holders must refuse, lest their entire stake be worthless.
That's an interesting point that leads me to the following thoughts: It's irrational to assume an attack scenario with only one single attacker. Instead, it's more reasonable to consider an unbounded number of potential attackers who could act one after another or even at the same time, without knowing of one another. So, the attack scenario is like a Poisson process.
For any distributed system that can only be attacked by somone who has a substantial amount (e.g. 1/2+ or 1/3+, 1/4+) of whatever resources (hashrate/power/stake), the security will depend on the time needed to carry out the attack (or more precisely: the time required to build up and possess the resources). The longer this timeframe, the more probable it is that other (non-colluding) attackers might launch their own attacks against the system in the meantime. However, when several attackers are acting at the same time independently, it is less likely that any one of them will have the power to not only outpace the honest nodes but all the other attackers as well.
Thus, making that timeframe longer will also make the system more secure. Furthermore, with growing popularity of the system, the number of potential attackers will also increase which further reduces its vulnerability.
Note that the attack timeframe is very short in Bitcoin at around 60 mintues that are needed for a transaction to be confirmed. PoW-PoS-hybrid or dual blockchain architectures can have much longer timeframes though.