Pattern matching has been successfully applied to phone logs between seemingly randomized src and dst phones. I'm busy at work and terrible at communicating clearly but what I am trying to say is that even when person A is using multiple phones and person B is also using multiple phones, there is still enough of a pattern involved such as call lengths, time of calls, etc to be able to sniff out and match these seemingly random devices to specific users.
I would have to dig up papers/reports on times this has been successfully done and reported publically.
But my point is that even running shit through coin mixers etc there is a high probability of there /still/ being enough latent information floating around to suss out a track between src and dst of blockchain transactions.