You're still approaching it probabilistically - a thousand factors can fingerprint the data and let our hypothetical attacker extrapolate (eg. output ordering). The more complication you layer on top the more likely it is to have cascading failures.
OK...perhaps you can give us, say, the top-ten of those factors that can fingerprint the data.
Also could you please explain how it is trivial to extrapolate the data when you control 50% of the MNs in your example above.
You're misunderstanding. If you are only ever going to perform a single transaction then the attacker only has one chance of observing it. However, you are (presumably) going to receive funds on more than one occasion and use Darksend to pre-mix them. The danger is greater for a seller, who will do this more often, than for an individual who gets paid his salary once-a-month. For both, though, our hypothetical doesn't only have one chance to catch you, they have an indefinite number of chances. And you can take it as a given that the number of MasterNodes they control will grow over time, not decrease, especially if it gets used for illicit activity by a portion of the userbase.
OK, I'm still not sure this is a fair representation. Hopefully someone who knows more about Darksend can chime in. I'll do some more research in the meantime.
And no offence, but you got it wrong about Darksend yesterday so I'm not sure that your assessment can be considered reliable.