If most of the miners collude like that than ETH will lose most of its value. The game theoretic aspect of PoS is an important part of its overall security.

This is only true if ETH remains a niche security and not a widespread currency. A large population won't drop a commonly used currency because of ideological concerns like miner collusion. Indeed, they'll have massive incentive to keep using it since their saving may be denominated in it.

The problem is not the same. In proof of work, it just breaks and you can double spend. In this proof of stake setup, you can double spend, but then the ethereum foundation can punish you with its slashers.

I don't understand this comment: "but staking 1eth is always staking 1eth".

If I have 100 GPUs and I buy one more can't I turn around and say "But buying one more GPU is just buying one more GPU"? Even if your point is that electricity bills are lower per additional GPU, surely humans discount the value of incremental money after some point so for those with lots of eth incremental eth isn't worth as much?

It’s cheaper to run 1 more GPU when you are already running 100

It’s the same price to add 1 eth when you own 100 eth as when you own 1 eth

Kind of, but doesn't that assume a linear utility function for holding eth which I think seems unrealistic? Plus, there are costs associated with holding eth (cold storage etc) which seem fixed, so its cheaper to securely stake 1 additional eth when you're already staking 100eth than when you're not running any. Plus fixed costs of actually running a staking business (by analogy with a mining business).

Yes, the utility function is linear for PoS, and wildly non linear for PoW