The current state size is ~5 GB. 500 GB seems like the largest that would be feasible in the medium term without requiring almost everyone to get special-purpose hard drives. Additionally, it corresponds to something like ~5 days of fast sync time assuming a 1 MB/s download bandwidth, which seems very inconvenient but not impossible.
Why constant? Storage technology is very far from physical limits.
It could be hard-forked up if needed, or adjusted up like gas limits.
Every contract would have to design some mechanism to collect fees from users (eg. token holders) to pay for storage. What if someone doesn’t pay? Either their data, potentially worth millions, gets deleted, or a fraction of tokens somehow gets sold. Sold how? Auction? How long and does that even work for something illiquid? So much complexity.
Users would automatically pre-fill the contracts that store any data relevant to them with a few years of storage whenever they send a transaction related to them. There are second-layer markets that would be required here to improve performance further, but they are simpler than the second-layer markets that would be required to maintain an acceptable quality of developer and user experience in a no-rent stateless-client-only model.
With an eternally locked eth model, lost interest from staking already creates an implicit time based cost.
Sure, but it also makes spamming the state cost-free for anyone who happens to be holding ETH, and will likely lead to the emergence of second-layer rent markets anyway. Heck, the status quo is incentivizing the emergence of second-layer rent markets, see http://gastoken.io.