This is an interesting point. I think the answer is that they are training and testing their tool to recognise individuals by picking up on subtle characteristic patterns in their writing. What those patterns actually are shouldn't affect how good the tool is at doing this, as the tool is trying to locate an individual against a backdrop of noise.
In reality, there may well be a problem here. It might turn out for example that MN users (and women in general) have greater variety in writing styles, or something like that.
They do have another three undisclosed forums and these may be of a dramatically different user base, so maybe the idea is to test the tool across different user bases to see if it works across demographis.
I'm reminded of the film Dances with Wolves, in which the actors playing the Lakota tribesmen were taught by a Lakota woman. Apparently the Lakota language has a female and male variant, so that the big strong male warriors ended up talking the female way.
Edit: big strong warriors, not worriers... Although maybe they were worriers too. If MN has taught me anything it's that being a worrier is not incompatible with being a warrior.