Isn't that the point with these projects? Managing a central service for 500 million users is hard. But with a decentralized setup, I won't need to connect to a million nodes. I don't know quite that many people.

Course, a friend request changes from a simple button click to getting my local couchdb instance to do a filtered sync with your couchdb instance.

You might have a point there. Although some people on Twitter have 100000 Followers and more.

I think if you have 1million active followers some kind of push publishing setup could help. But as far as I know there's nothing in the tech of couchdb that prevents replicating to a million other nodes.

One obstacle is that you cannot have one million open sockets (for long polling) at the same time on one node. I don't know how easy this is to mitigate.

You can have a million open sockets in Erlang, although it requires serious RAM: http://www.metabrew.com/article/a-million-user-comet-applica...

Hmm. The changes are posted through comet in continuous replication? That would mean another distribution system would be better for heavily published nodes.