I have an application that transmits data once per minute, and I chart the RSSI. I've seen it occasionally as low as -90dBm, maybe even a couple less.
It seems to work down to -90db but the packets are only about 20-30%.
Interesting that during the night when the house is quiet (electrically as well as otherwise), signal strength is stronger and very consistent. When I'm in my office where the one node is, with two or three PCs running, and lots of fluorescent lights with electronic ballasts, things bounce around much more.
When I get it near the computers and other rf noisy devices, the measurement does indeed drop. It seems to be a measure of usable signal not raw signal strength.
The other interesting thing (I'm using S2 XBees) is adding a third node physically somewhere between the other two. This will sometimes cause an increase in signal strength because the traffic will sometimes switch to be routed through the intermediate node. So that's one way that the mesh networking can be seen doing its thing
The RSSI for the intermediate and the furthest nodes will then be nearly identical -- RSSI only reports the last hop.
That could be useful for range extension. I am new to working with XBees and working with Series 1 since they are alittle easier to work with for newbies.