Most of my opinions about nostr's architecture come from a conversation I had with a friend of mine who used to work at Twitter. He explained that Twitter is only able to deliver relevant content quickly by pushing it to a huge network of special purpose caches. This is easily done with nostr relays in theory, the hard part is coordinating how caches (relays) get primed. I think some combination of push and pull will be necessary - if indeed the problem can be solved without central coordination.