MAN as an Overlay Network
One use for a MetropolitanAreaNetwork is to provide existing network (Internet or private) users to communicate with each other without using paid-for infrastructure such as T1's or cable modems. A wireless mesh of sufficient density and interconnectedness allows two users in different parts of the city to share data without touching the public Internet, and potentially do so at higher speeds.
However, the resources that a given user can provide themselves are rather limited and generally not of interest to a large number of people. Even if Google were to directly connect to the cloud, allowing hyper-fast local access to www.google.com, the total benefit would be essentially zero, because the whole point of the Internet is to hyperlink to other sites.
The resource that every MAN user can provide that we all want is connectivity. This is not just for access, but also for transit. Google may not be the best thing to be directly connected to the cloud (though since it's also a cache I disagree that the value of that would be near zero) for the link-related reasons you mention. A content provider like a large news site might be a different matter (suddenly everyone would be able to watch the video clips). For myself, most of the people I know are local. These are the people I'm most likely to want to send a video to, or allow to drop a huge tarball on my server. The ability to do all this at 10M rather than 128k (my DSL uplink speed) is a clear win. This would sure be a nice way to telecommute. ---ScottPeterson
The only resource that's useful in a general sense that nearly anyone (with hardware and the right config) can provide is a cache. This includes both web caches (think Squid) and mirrors of large archives (think Debian). With the right software distributed across all the servers in the cloud, web surfing and site mirroring could be made to operate in parallel, load-balancing each of the Internet connections available to the cloud.
While I've also been interested in collaborative web caching, I've never figured out how to adequately address the privacy concerns (e.g. do you really want your surfing habits recorded in your neighbors' web logs). We may have to resort to something like what Zero Knowledge Systems was doing to mask the origin of web traffic (this product seems to have disappeared now). A related application that can also benefit from pockets of highly connected nodes is distributed filesystems like freenet (freenetproject.org).
One thing a MAN might be good at is distributing multicast streams. There seem to be two obvious benefits to this. The first is efficiency. If two participants want to watch to the same internet TV station (assuming it's using a multicast stream) one of them can relay the stream to the other over the MAN, and the second participant's bandwidth to the Internet can be used for something else. The second is access. According to multicasttech.com only 5% of ISPs support multicast. [Rant: If I were a big media company that owned a broadband ISP, I wouldn't be in much of a hurry to provide multicast service either, since it will just enable my customers to buy content from some other provider.] A MAN that supported multicast and contained some participants that can receive multicast traffic can obviously distribute some multicast streams to other MAN participants that can't get the streams any other way. How the participants with multicast gateway capability decide which streams to spend their contributed bandwidth on is TBD.' ---ScottPeterson'