The vast majority of people that show 90% or more correlation with me are concentrated in 2 areas of the world, New York city and California (SF Bay in particular)

I tried the same experiment and got the same results, modulo London. That seemed odd to me, so I punched in a couple of smaller cities where I've met reasonably interesting people and asked for match results back. Sure enough, it returned a few screens of 90+% matches in both cases -- comparable to the "everywhere" results, in fact.

Although I don't feel like putting enough effort into this to figure out exactly what's going on, it seems clear that the "everywhere" query isn't searching the entire database. I'm located in the SF Bay Area, so I'm guessing local results are overrepresented in what I'm getting back; it seems excessively convenient that I'd just happen to be living in my best possible social scene. NY and London are both among the largest English-speaking cities, though, so not much surprise there.

If I'm right, it'd be hard to use these results to gauge social compatibility unless you could correct for the missing matches. If it's dropping some geographically or demographically asymmetrical subset of possible matches, it might not be possible at all.