Actually, the number of questions you get wrong is not the only thing taken into consideration when the software calculates your score. The difficulty of problems you get wrong and the way in which they're distributed (i.e. a string of 5 errors is much more severe than 5 errors spread across the 37 questions).

The fact that you got a 46 in quant with 17 errors but a 38 in verbal with 9 errors should not surprise you. Remember, it's not the numbers themselves (46 and 38 in this case) that matter, but their respective percentile ranking. A 38 in verbal is equivalent to 83% (you did better than 83% of test takers), while a 46 means a 76% (you did better than 76% of test takers). As you can see, these percentile rankings do follow to some extent the number of questions you got wrong!