Of course they do, but it's not as big of a deal as you think. Anywhere between a 38 and a 42 is extremely variable, and anywhere above a 33 is going to be competitive at any school in the nation other than maybe the top 5 stat-mongering schools. But think of it this way: Last year about 1300 students got a 38 or higher on the MCAT. This is enough to fill up the top 10 schools in the nation. However, only one school even has a
median of 38, and that's the notorious Washington University in St. Louis. Few schools have medians in the 37s, and a majority of the top tiers have medians between 34 and 36. Clearly, these schools could fill the classes with 38s, but they choose not to for many reasons. One of which is that the differences between a 34 and a 38 may not be as drastic as say, their personalities and other accomplishments, and Admissions Committees know this.
There isn't a huge amount of difference when you're speaking in the 90th percentile. Once you're in the 36, 37... 42 range, you're dicing up individual questions. This is why schools don't look into the higher scores as much. The real weed-out comes in the low 30s and high 20s range. This is because the test is very forgiving around section scores of 8, 9, or 10 where point spreads can be quite several questions, whereas the score differential at 13, 14, and 15 is generally only one, two questions maximum.
Overall, the scores represent percentiles, not a consistent scalar quantity. The numerical scores are used to more quickly and easily review applicants, though percentiles are more accurate. That's why this system exists.
I will use the 2011 data as my example. However, all the data for the past 12 years can be found
here at the AAMC website.
If you look at the E-MCAT score spreadsheet that the AAMC offers, the difference between a 37 and a 40 is as little as four questions, and only a max of 2.5% of all test takers score in this range. Now, if you look at a 37 to a 34, up to 7% of test takers score in this range, and the difference in questions right between these test takers could be up as far as the ten range. Obviously, this isn't a huge difference, but you're currently looking at the > 90th percentile of test takers only (34 or greater is 90th percentile).
Now, the big differences come around the 30 range.
Each point in this range is about 6% of all test takers, that's more than all test takers scoring >34! Also, the number of questions difference in a 4 point spread doesn't just indicate that about 25% of the test takers fall into this range, it also indicates a significantly larger number of questions wrong (ie lack of knowledge) than the very high or very low score ranges. Just like all standardized exams, the test is designed to designate who can comprehend and can also demonstrate their knowledge in a standardized testing environment (which you will face many, many more times in medical school, residency, and beyond). The focus is on distinguishing the middle of the pack, not either end of it. The single test design may be a poor judge of people scoring < 16 and > 34, but is pretty accurate between those numbers.
For more detail regarding how the exam is scored,
you should view the AAMC's website regarding the topic.
Edit: I hope this is helpful and enlightening, but I'm going to put a disclaimer in here. I'm using my own research and interests in the topic (I got bored studying for the test so I found other things to do... haha) and what other reliable sources on SDN have said. If an AdCom were to come in here and say I'm full of ****, I would go with their insight, not mine.
