One thing I'd like to better understand about cMix is the practical implications for flow analysis attacks (watching who participates in each round and trying t

One thing I'd like to better understand about cMix is the practical implications for flow analysis attacks (watching who participates in each round and trying to correlate senders and receivers based on participation in the same rounds). cMix seems to accept more likelihood of a worst-case adversary in order to maximize efficiency. For example, the paper says: "Another notable difference between cMix and most previous mixnets is that each mix node knows all senders. This difference does not weaken the adversarial model because the adversary is expected to know all participants of the mixing round, and in cMix the unlinkability between a sender and a receiver is still ensured, by even any one uncorrupted mix node. On the other hand, this can empower cMix nodes to perform other tasks such as end-to-end secure messaging without introducing a public-key infrastructure of the participants." It's true that in theory mixnets are often modeled by assuming a perfect adversary who sees every message enter and exit the net and can perfectly match it to an ongoing participant over time. But I believe that they don't claim to offer perfect protection against such an adversary, and in practice benefit from mostly not having to. It seems like cMix makes it easy for a compromised Network Handler, or maybe any one compromised mix node(?), to serve as that perfect adversary. So ... does the design of cMix make it more likely that, in practice, it will have to actually face a worst-case adversary that most mix networks don't have to face?