It isn't true there is "no correlation." Obviously there are all sorts of correlative coinjoin attacks, timing attacks, sybil attacks, etc. That's why darksend does what it does with mixing ahead of time, multiple rounds, denominations, etc. -- to try to defeat these correlations over multiple rounds. However, there is really no way to know how many rounds are needed, how long to wait after receiving coins to start mixing them, how long after mixing to use them, etc. Are 3 rounds enough or do you need 300?

The mathematical analysis that is presented on this to give numbers like 0.0000001% is flawed and deliberately misleading. They are mostly treating as independent events things which may not be independent and also are assuming an ideal model with no leaks (even probabilistic) in each stage. I don't believe it is possible to accurately compute these numbers in a real-world adversarial context. And that's just for public blockchain analysis. Once you include additional information from masternode logs the problem gets even messier.

To be fair various combinatorial, timing, etc. attacks exist with cryptographically-private systems such as cryptonote too. The main difference is the lack of additional inputs (masternode logs) from untrustable third parties, and also the lack of a simultaneity requirement for mixing.