InspiAir says they can span kilometers with normal 802.11b in point-to-multi-point formations just via software: Yeah, and I remember Karlnet, too. Back about five years ago, Karlnet sold firmware upgrades for common reference designs, like Apple's AirPort Base Station, that would allow, wouldja figure, timing changes and other protocol tweaks that allowed 802.11b to span several kilometers in P2MP configuration without failing.
InspiAir seems to be implying that they have mesh technology, too, according to this Techworld account. I don't think so. One of the investors is quoted talking about the "not-yet-patented technology." Given prior art, I would guess that might be "never-to-be-patented technology."
I don't quite understand why they've made a splash. The "100 milliwatt" radio they talk about using would be the unamplified signal coming out from a card. Stick a sectorized antenna on that and work within point-to-multipoint rules and you're allowed (in the US at least) to push out pretty high effective wattages that can span many miles.
It is, in fact, a violation of the law physics if you were using 100 mW of actual output power measured at the antenna and expecting to reach a few kilometers. Or more than a few thousand feet line of sight in a non-interferring, non-reflective environment.
Such as in a park. In the Techworld article, an operator in New York which provides service in the Hudson River Park is quoted as saying that they were astounded by how two nodes from InspiAir could cover a three-quarter square mile park, where 25 to 50 nodes is often quoted for outdoor deployment for a square mile. Yes, but that's a square mile of dense usage with obstructions. It's not a big deal to use high-gain antennas cover 90 degrees of arc at opposite ends of a square to cover that range. That's only a few hundred feet from either transceiver. When the leaves come back, then we'll talk about coverage.