I've thought about this as well. There are some additional points to consider. GMAC counts number of tests taken, and not number of test-takers; so an individual could conceivably take the exam 12 times a year and get 99th percentile each time and single-handedly skew the results.
I don't, however, believe that if there are 2000 GMAT tutors worldwide, that they consistently score in the 99th percentile. There's a reason why MGAT & Veritas pay $100 an hour to their GMAT instructors, who must score in the 99th percentile. Kaplan and PR instructors must score in the 90th percentile, I believe, which is 10x more common.
I wouldn't be surprised of a few hundred scores a year, maybe 10-15% are taken by professional instructors and test-takers. You can probably throw in people that take the exam but do not apply to business schools, such as those going for PhD & MFE degrees. And then there are a limited number of people that score over the 95th percentile, but then re-take the test. Since scores are only used once per person, any additional scores factor into the testing numbers but not the application pool.