OP, in your scheme "cooperating" citizens don't give any more information to the "interpol" than if they were not part of the mix at all. That is, "interpol" could just as well ask them not to take part in the mix. The only difference they make is that people who want their coins anonymised have more uncertainty as to how many people are using the system for the same purpose.
I see what you're saying. But if you assume the criminals don't all know each other and group together (which they might well do) the statistics do imply more cooperating people is good:
Say we have 900 people who would cooperate, and 1 criminal, and 99 non coops.
Now if only 100 cooperating people use mixing,
the total mix pool is 100 coops + 100 non coops.
If we're doing 10 person mixes, on average we'll get 5 coops and 5 non coops in a mix. Tracing an individual criminal is now pretty hard - you'll get 5 leads per investigation.
But if 900 cooperating people use mixing, the total mix pool is 900 coops + 100 non coops, and on average there'll be only one lead per investigation.