Let's say a Tor user has a history of routinely accessing websites A, B, and C, and that this history is established before he began using Tor. Let's say some adversary, like his ISP or a government agency, can access this history after the fact. Suppose that the set {A, B, C} is enough to accurately fingerprint the user --- almost no one else routinely accesses that particular set of websites.
Further Suppose the adversary also has the ability to monitor some number of exit nodes. If the user routinely accesses websites A, B, C, and X using Tor, how difficult would it be for the adversary to conclude that this specific user is accessing site X (because all four websites tend to be requested within a short time frame)?
I'm not talking about a basic traffic analysis attack where the adversary actively monitors both entry and exit nodes; I'm talking about an adversary who has a partial behavior profile on the user and can monitor exit nodes.