Actually, your performance on GMATPrep isn't strange at all, at least acording to my GPrep experience, which is... well, lets just say I took tests 10 - 15 times... Each.
With 3 incorrects in the first 27 questions the test has "put" you in the "51" bracket. Had you continued with such a performance you would have ended with 51 in the end of the math section. But with 12 minutes left you couldn't and, due to random guessing on the left 10 questions, you ended up with Q47. I would say it is a perfectly normal score for such a test pattern.
If you repeat that on real test, with only 3 incorrects in the first 27 questions (especially if incorrects are clustered near 20-27 group) I am 100% sure you will end up with 45+ raw score regardless of your performance in the last 10 questions.
In addition,
MGMAT have conducted some analysis of the Gprep software in which they followed the results depending on the number of incorrect answers in the various segments of the test. They answered randomly on the first ten or the last ten questions with almost 100% performance on the rest of the test. When they blind guessed the last ten questions they encountered the scor drop very similar to yours.
Unfortunately, all of this just confirms the old myth that 10 to 15 questions are extremely important on the GMAT and, consequently, greater attention should be given to them.