That would solve a real, current problem and save consumers money, instead of trying to preempt the highly speculative possibility that our government would want to shut down the internet.
Also, just local wireless mesh has plenty of technical challenges. As other posters have noted, significant challenges stem from both sparsely populated areas (link distance) and densely populated ones (interference, spectrum limitations). There has been significant research into overcoming these limitations--see, for example, MIT's RoofNet and UIUC's CuWIN.
One project I think is especially interesting is Phil Levis' research at Stanford, studying full-duplex wireless. For the first time, they have nodes that can receive and transmit on the same channel at the same time. This (almost) doubles max throughput, but more importantly, it solves a lot of difficult interference problems (particularly the "hidden node" problem: http://en.wikipedia.org/wiki/Hidden_node_problem).
There's also a hard, unsolved software problem: how to route traffic effectively on a large mesh of unreliable, low-throughput routers. No existing routing protocol that I know of (RIP, OSPF, BGP, etc) is well-suited for this.
I would love to see a startup built around last-mile wireless. They could, for example, use commodity PicoStations along with high-gain antennas. The hard part would be nailing the business model and the software. Both are fascinating problems.