If the goal is the sort blobs into buckets for performance reasons then what about taking the first 8 char of the hash as the "directory". should be perfectly random.

If the idea is to allow users to sort blobs into "folders" and have filenames, then I would caution against storing too much metadata that couldn't be transferred. blossom servers are in the business of storing and serving raw binary data. nostr is the protocol for storing metadata

For example there is nothing stopping you from making a plugin for NextCloud to support blossom endpoints. but users are going to be very confused that their "decentralized", "censorship resistant" files dont have the same names or folder structure on another server.

Either way I don't see any harm in servers having their own special upload endpoints or GET endpoints. just so long as they support the core BUD-01 and optionally BUD-02 it should be enough for users to effectively use multiple servers without worrying about each of their special features

I think of it like relays, some support negentrapy, but all support NIP-01 filters. so all clients use filters :)