Actually, there is quite a bit one can derive just by timing the requests and payload size. Given that the size of wikipedia articles at any given time can be calculated, as well as articles they link to, it's quite possible one could reconstruct a given wikipedia browsing session using metadata alone.