* Media itself is not stored, only links to media.
* Relays are free to limit usage, charge a fee or require proof of work.
* There are many kinds of messages with different storage standards i.e ephemeral messages are not stored at all.
I'm not an expert but the SSB scaling issues seem related to the architecture being tied to the sigchain, & general feed immutability. This makes it very difficult for pubs to, for example, accept/process subsets of posts or do anything out-of-order. It also makes it impossible for clients to limit local database storage, especially mobile clients where that might be at a premium.
Even just the existence of ephemeral notes seems to indicate there's no such architectural limitation so that sounds really hopeful.