I used to have the same kind of impression, but I don't think you can accurately predict the raw/scaled scores basing on just the number of correct answers.
I have taken enough gmatprep tests
to confidently say that the "adaptive" nature of the test makes it difficult to determine the scores basing on just the number of wrong questions.
I think the key is to string together as many correct answers in a row as possible. The consecutive correct answers will determine your upper limit, once you hit a high upper limit, you can get wrong answers and your score won't go down as much. But if you get a wrong answer for every 3 to 5 questions interval, it is more difficult for the score to increase.
For example, I have had 6 wrong questions on the math section and got a 51, but 4 wrong questions and got a 50. For the 51 score, I got the first question wrong, and another wrong answer about the 20th question, and 4 wrong answers in the 30-37th questions. But for the 50 score, my 3 wrong answers were concentrated in the 20-30th question range and the last wrong question was question #34.
As for the verbal section, I would get 8-9 questions wrong and get a 40-41 score, OR 8 questions wrong and get a 43-44 score. The difference is that with the 41 score, 7 out of 9 wrong answers were sentence correction types, whereas the 44 scores 5 wrong answers were SC types, and the other wrong answers were spread out among the CR and RC sections. My impression is that (if you can control the absolute number of wrong questions) it is better to spread out your wrong answers among the sections instead of having them concentrated in 1 section, BUT we don't have that choice.