Hi gmatpapa,
The free diagnostic test that you took was not a CAT. The scores are generally accurate, but the free test is designed more to familiarize students with what they do and don't know in terms of GMAT content than to provide a precise score.
As for the odd percentiles, it IS mathematically possible. For instance, if A, B, C, and D respectively scored 1, 2, 3, and 5 on the math and 5, 3, 2, and 1 on the verbal, then B would be 2nd in math, 3rd in verbal, but tied for last in overall points. Of course, that scenario seems fairly unlikely; please PM me your name and email, and we'll investigate your specific case.
Finally, I wish I could answer your last question, but I'm not clear it's the case. I just posted in another thread asking why our tests were too EASY! Practice tests are inevitably unreliable; beyond the fact that no one in the entire test prep business can spend the time and money on each question that the GMAT does, the psychological pressures and motivations that a test-taker faces are completely different on the real thing. What we do know, however, is that our own surveys show our practice tests are an accurate predictor of test-day performance, even if students do occasionally report scores that they feel aren't representative of their normal capabilities.
Please let me know if I can answer any other questions for you!