No it won't. REQs will only have specific authors so tons less bandwidth for more connections. This removes the issue of downloading duplicate notes which is what really kills bandwidth and relay compute . And users don't have to manage any relays 

 True but I still don't see how connecting to a ton of smaller relays from a client will result in user experience matching/exceeding centralized services. My hunch is this will have to be solved server side and with trust 

 I think eventually main way to consume notes will be using relay feeds like what we're working on at wss://algo.utxo.one (you can try at https://jouble.surge.sh/ on desktop)

Pick specific algorithmic feeds vs simply composing chronological follow feeds 

 Awesome! And yes, totally agree