Whats 'many'? 5 nodes, 10 nodes, 1%, 10%, 20% of the nodes, more? Based on your calculations, how many dishonest masternodes does it take for transactions to becomes less than 'even some modest degree of privacy'?
There are different vectors of attack here. If we ignore inherent coinjoin-type issues (the sorts of things this thread was originally about, which may be a false alarm, but those sorts of issues still exist in general terms and may apply to DRK in some ways), then the numbers I'm talking about involve most of the masternodes being malicious or compromised for catastrophic failure to occur. I don't believe that is the case today; I think most masternodes are run by people who either support the DRK project or are just doing it for the money (the latter is fine, short term, as it means they aren't interested in compromising privacy).
But let's face it, what is going on now doesn't really matter at all. The whole point of this exercise is to build something secure for the hypothetical future where these technologies are very widely used and important. That is the point where I do not expect the current degree of masternode fidelity to continue, or at least I find the blind trust that it will irrational and unacceptable.
That said, even in the case where most (but not all) masternodes are honest, at least with the current system, some transactions will be vulnerable. With 3 rounds chosen randomly and 2% of masternodes being dishonest, it means one out 100 000 transactions will be completely unmixed (more will have reduced anonymity). That may sound great, unless you are that one. Still, this might be an acceptable degree of risk, if there weren't a better way.
There is simply no substitute for strong cryptography that does not rely on third parties for your privacy. If there is no feasible cryptographic way to do what needs to be done, sure build trusted third parties and try to minimize that trust by working through several of them. But there is a feasible way to do this without any third party trust at all, which makes the masternode approach unnecessarily dangerous.
This really applies only to anonymity and similar cryptographic applications though. If you want to use masternodes to validate instant payments or provide other services that don't have the same trust requirements as anonymity, that might make a lot more sense.