They can be, and that's a statement about the distribution and consistency of skill in those activities. I see little reason to believe those distributions apply to SAT taking or general capacity to learn (which is what you're really aspiring to measure).
If we set up an ELO league for best-of-three rock paper scissors, I'd expect score to diverge over sufficiently many hours, but would give almost even odds to a 99th v 93rd percentile candidate.
Sure, that may be true of ELO-based games. In the realm of the standardized tests used for school admission, we also have a lot of data about performance so we can use that data instead of looking for analogues in other fields. And my experience with the SAT & ACT leads me to believe that on these tests, the difference between 1% and 7% could be something as simple as how well the student slept the night before the test.
Skill difference between top 1% and top 7% is huge in ELO based games