Very long story short, the overhead in maintaining the link state information in meshed networks swamps the available bandwidth for even modest (think 20-40) numbers of nodes. Basically, as nodes are constantly moving/coming on- and off-line, nodes have to know which nodes are in the connection graph, so they exchange that information. But that eats up all of the bandwidth amazingly quickly, leaving no room for actual traffic.

There are some other issues with meshed networks (latency, battery usage when you're just a client node, etc.), but especially for those of the kind described/dreamed about here (which are also known as MANETS -- mobile ad-hoc networks), the routing issue is a huge one.

Finally, I should note the above isn't just academic; several companies have tried to build products based on these kinds of networks -- military contractors being one of note -- and have pretty much come up short.

So until the really fundamental issues are worked out, which is in the realm of academic research, not implementation, we probably aren't going to see anything like this.