In Ethereum there's one disincentive to stake pooling: if a staker misbehaves, perhaps because it got hacked, the penalty depends on how much stake misbehaves at once. A large amount of misbehaving stake could be completely destroyed, while a small amount will lose only a small portion.

This is mainly because a small amount does no damage and is presumed to be accidental, but also helps decentralize since you're safest not using a large staking pool, the most popular client, the most popular hosting service, etc.

Running as a small staker is feasible because the minimum stake is only 32 ETH, you're profitable if you're online at least 2/3 of the time, and the load on an individual node is feasible for a laptop. The computational load increases fairly linearly with the amount of stake, so there's little economy of scale for a large staker.

There are two attractors, it could push people to do it themselves but I think it's more likely that it evolves in stake pooling. Stake pooling in pure PoS involving trusting the pool operator to a great deal and I'm concerned it degenerates to a very centralized system. Adding optional delegation is a safety valve that takes this into consideration.

What Tezos does is merely let you lose a different key to custody funds and to create blocks. By itself this creates the possibility of delegation.

Last but not least, with trusted hardware the difference blurs even more.

You need to control 2/3rds of the stake and be able to coordinate decisions with each staking node before the block time is over to fuck up the network. Considering that PoS can enable more tunable block times as validators no longer will churn out random numbers on GPUs/ASICs to mine a new block, I think this will be a non-issue.

Unless they somehow integrate PoS in the wallets, yes.