I wonder if this isn't susceptible to statistical methods. The idea would be to both issue requests and to operate onion nodes, feed a big number of requests fo

I wonder if this isn't susceptible to statistical methods. The idea would be to both issue requests and to operate onion nodes, feed a big number of requests for the Silk Road into Tor, use timing and size information for the requests at your nodes to identify those that are possibly yours. Every request to the Silk Road has to terminate there, so the IP-numbers that it uses should sum to 1/(# of onion layers) of your total, minus a factor for packets you falsely identified as yours.