It sounds like a simple question, but I think the answer is a bit complicated. When you take your GMAT, you aren't directly compared against other GMAT test-takers at all; you're compared against GMAT questions, which in turn have been calibrated by earlier test-takers.

As I'm sure you're aware, each GMAT test includes several 'diagnostic questions' which do not contribute to your score. They are included so their difficulty can be gauged for future tests. This process is a bit complicated, so I'll simplify a lot:

-if a diagnostic question is not answered well by most test-takers, it might be calibrated as a 90th percentile question, for example;

-if this question is then included as a real GMAT question a year later, and the test-taking population has improved substantially, it may be answered better this time than it was when it was used as a diagnostic. That is, more people than expected will get the benefit of answering this difficult question correctly- the test effectively says to itself "this correct answer gives me some evidence that this test-taker is above the 90th percentile level". This will boost scores, overall, compared to earlier populations;

-the upshot: you're being compared against the population that had your GMAT questions on their test- when your questions were used as diagnostic questions. I don't think there is any published information about how long the interval can be between when a question is used as a diagnostic and when it is used as a real question.

Of course, many test design details are proprietary, and anyone answering this type of question, unless they work as a GMAT test designer, must make some educated guesses, as I've done.

While I find a lot of these details interesting, I don't think the GMAT test-taker gets any advantage from knowing the scoring algorithm and calibration process inside-out. You want to know enough to devise an intelligent pacing strategy, but that's pretty much it.