From a dumb layman's perspective (mine), it seems as though the addressing issue has a very simple solution: to refer to some arbitrary file which is hosted by some known third-party (say, a video on YouTube), provide a link to it from your server and distribute the link to that link to others. When switching from YouTube to Vimeo, simply change the one link on your server and the link you distributed remains valid.

This also avoids the danger of hash collisions, although it is vulnerable to third-party hosts taking down the content (although again, there's not much cost to that. Just move the content elsewhere and change your redirect link).

With this method you do still have some server responsible for hosting the content rather than distributing that load, but that's actually a saving when looking at the storage load across the entire ecosystem.

It's a cool article and I really like the idea of content addressing instead of host addressing. I just feel like it's too divergent from the web as most people use it today (and I'm mildly concerned about malicious hash shenanigans), whereas the above method can be used and understood right now by many web users.