Justin Chapweske steps up to bat with a draft idea he calls "the content addressable web". The core idea is to improve the experience of getting big things from the web by naming resources by pointers to the resource, not the resource itself. Then you can have a transparent way to mirror resources. For an added bonus, those pointers can include secure hashes of the contents, so you know you got the correct data. Using URIs this way isn't entirely a new idea, but Justin's version is good.
I'm concerned that we'll never get to a web using "better URIs" to identify resources. We've been running around this idea for eight years, and still nothing. Justin's approach has the virtue of being simple and incremental. For more, see the discussion on the decentralization list.