The major problem with this proposal is scalability. "Repeater mode" is usually WDS or a logical client bridge+AP. This is entirely layer 2, has no routing, n

The major problem with this proposal is scalability. "Repeater mode" is usually WDS or a logical client bridge+AP. This is entirely layer 2, has no routing, no useful path metrics, and is completely unsuited for a network larger than a few nodes if you want any sort of reliability/usability. True meshing algorithms like 802.11s or BATMAN can optimize the mesh topology, but you're still on one huge broadcast domain. The proposal needs to establish a means to restrict the size of these domains - convert each node (or a subset of supernodes) into "routers", and roll out a routing protocol which is optimized to this type of ad-hoc, organic deployment. The other problem with scaling (no matter what your meshing algorithm does) is the half-duplex nature of RF - if you're doing this with single-radio SOHO routers, you're going to lose half of your throughput per hop. I run two 500-node muni WiFi nets, and we try never to exceed 2 hops between a node and its backhaul gateway - any more and you're in the sub-1Mbps range on 802.11g rates. Fix these problems and you may be on to something. Of course you also need to bridge geographic gaps somehow.