I think it’s practically impossible to obtain the data for offchain reads, even if we work with the RPC providers to get those data. For example, I could run an RPC node myself and read the state that I want.
Regarding the temporary storage model, it depends on what it’s used for. For example, if I want some data to be persisted onchain forever, then use permanent storage. If there’s some data that I’d like to persist for let’s say maximum 6 months or 1 year and I do not care that it’s gone after that, then use the temporary storage.
Regarding why people would use temporary storage, the obvious incentive is that it’d be cheaper. But whether people would actually use it (i.e. if there’s some data that can be stored temporarily onchain and not permanent), we have to do more investigations.