Here’s another use case: things like EEZ that require an execution table to be prepared in order to serve pre-proven rollup calls. (See Synchronous Composability Between Rollups via Realtime Proving on Ethresear.ch) After a few transactions at most, the table can be discarded, yet discarding it actually costs more gas than leaving it laying around, which is a bit backwards, like fining people for picking up their trash. That said they might be able to make do with transient storage, more research needed.
There’s an interaction between this EIP and EIP-7928: Block-Level Access Lists which I think should be pointed out: block-scoped storage needs to be added to the block-level access lists, otherwise we lose the ability to parallelize execution. EIP-7928 will be on Mainnet this year in Glamsterdam. As a result I would suggest raising gas costs of block-scoped storage closer to, but not exactly at the level of persistent storage.