There are different tradeoffs when you're talking about a mesh based on physical topology versus the logical mesh of a DHT.

I think we'll see some rapid advancements as Van Jacobson's NDN[0] and Miller's telehash[1] reach maturity.

[0] http://named-data.net/

[1] https://github.com/telehash/telehash.org/blob/master/backgro...