Good point. I wonder how useful that is in practice with the amount of traffic going through the Tor network. It seems to me that the more people use it, the harder it would be to get accurate correlations. That said, I wouldn't be surprised if some clever math can do so more accurately than has been published.
A correlation attack[1] doesn't care about the intermediate routers. It only requires packet dumps from the entry and the exit node. Both of which, with very high probability, route through networks that are monitored by the NSA.
[1] https://blog.torproject.org/blog/traffic-correlation-using-n...