Some general observations about scoring discrepancies, probably not very pertinent to anyone but worth pondering over by those test takers who go through that experience.
MGMAT's (and that of all the other prep companies) scoring algorithm can be VERY off sometimes, depending on factors such as the company's own (in-house) interpretation of question difficulty, sequence of errors in a row, or the question number at the which the errors are made.
Since PearsonVUE/GMAC pretty much don't reveal the exact rubrics of their scoring algorithm or CAT systems, the huge discrepancies between practice scores (be it MGMAT or some other prep company, or even the GMATPrep on the 2nd and 3rd runs of the software) and the actual test are not actually very surprising. These prep companies are trying to guesstimate your testing level to the best of their abilities and probably doing so with a bigger standard deviation than they might be willing to accept.
In reality, score discrepancies often indicate that the test taker was not at the scoring level that the practice tests were suggesting and that some fundamental GMAT concepts are missing from the preparation.
In other words, unless the test taker was sick or nervous or had some psychological issues or even timing issues, the score that he/she receives on the real test is a reflection of his/her GMAT level. Of course, luck (guessing well on that day or getting fewer questions from your weak areas and more from your fortes) can play a part on test day and can be the difference between a 680 and a 700, or a 750 and a 770, or occasionally a 680 and a 720, but if you are scoring in the 740s on practice tests but manage to pull of a 650 on the real test, chances are that you are not yet a 700-level; the good news is that, with good practice you can surely get over that hump, given the potential that you have shown on your practice tests.
On another note, mid to upper-700 scores on second and third running of the GMAT Prep software should be disregarded. Subsequent runs of the software should be used purely for practice and not for estimating scores, because of possible MASSIVE score inflation from points gained and time saved in picking an answer that one has already seen in a previous run of the software.
Good luck, folks, and keep digging in, sooner than later you'll get the score you want (or that +/- 20, depending on how the wind blows that day and the mood of the GMAT gods).
Der alte Fritz.
+1 Kudos me - I'm half Irish, half Prussian.