pull down to refresh

This is very similar to the theoretical idea that in bitcoin the big mining pools will take the mining hostage and only build on top of each others blocks and censor small miners.
This is more related to the proactive vs reactive security, IMO. Since performing this attack on Bitcoin would require much more than 51% of the hashrate to pull off, plus the direct cost of maintaining the attack.
In your scenario the big pos stakers would take the staking hostage and not validate what small validators (=non exchanges) do.
More realistically it would immediately create a fork in PoS since I think they're more fragile to network partitions.
More realistically it would immediately create a fork in PoS
Agreed. Pow incentivizes time/energy spent on the thing that one already did spent time/energy on (which is also why there will never be a 1:1 copy of bitcoin). Pos does not have this natural self incentive. In any ethereum fork people can just stake on both forks which is detrimental with ethereums centralized foundation and a demi-god vitalik.