I don't see why not. Is the problem fundamentally different than SMS messaging or email?
- How would users know which nodes they could trust?
I think email and Sms are pretty weird user experiences that have succeeded in part because there was no alternative.
That is what search engines are for. The whole point of the Internet is that you can build businesses on top of open protocols with interoperable implementations. Giving up on open protocols and control of data (you can't even save images on Instagram...) "because hashtags" is incredibly shortsighted.
Perhaps I was wrong to say that federation is in principle ruled out, but I do think that it must be invisible to users. - Including namespacing usernames to nodes within the federation as with the @ sign in email.
I believe this is true in spite of the technical beauty of federated solutions, and it's exactly that beauty that's lured so many onto the rocks.
As soon as a user has to understand federation - at all - I believe the proposition is dead to most people. For example Diaspora's 'pods' are, I think, fatally confusing. (I think Zuckerberg realised this, which is why he donated money to Diaspora - a project he could see would fail, but might also crowd out other OS alternatives to Facebook.)
For sure, XMPP can be part of the tech solution, in fact I suspect there are really excellent open source tools to address many of the inherent issues (perhaps Apache Cassandra can be Twitter's contribution to democracy) but I think the real problem is a social one.
You need an institution that can credibly promise to provide a platform in the long run and which can also get critical mass in at least one community. I believe the critical mass might well follow if the institution is widely perceived as reliable and legitimate in the way a foundation like Wikimedia is.
I went into some more detail on this in a blogpost: http://jimmytidey.co.uk/blog/building-a-better-twitter/