"a sort of virtual-space map, which would reveal nothing about actual phyical location of the sites or the person you are contacting."
I'm not so sure... in cellular systems, cells must know where the handsets are located in order to send incoming calls. Your transmitter has a physical location which could presumably be tracked in the normal manner, and I would expect the overall routing information in a net to be susceptible to traffic analysis in any case. An individual who is using the system to communicate wouldn't be able to find the physical address of another user, but e.g. an intelligence agency which was looking at the entire network would.
You've hit the key concept - motion. If you are operating from a fixed transmitter, finding locations is very difficult. The problem is that if someone was to drive around the city in a radio equipped truck, and log into lots of different nodes, they could at least get a partial map of where various nodes are located. This could be countered by additional software which would emulate the node you logged in on even tho it actually switched you to another node (you communicate with a virtual fixed node which may or may not be the actual one you are communicating with). hmm... Which leaves us with the problem of developing software to do that, and developing some method of logging into the system which would not be node-specific (if you had to do something special for each node, it would immediately reveal what areas the nodes were in.) That creates the problem of developing something that is not a plainly obvious "log-in" signal that the FCC could look for. Ideas?