I think I saw an attempt at trying to optimize filters in the NIP PRs, but I don't know how far it went.