I think the math doesn't work out the way you're looking at it.
First, yes, most talent searches use a 95th percentile cutoff. But that doesn't mean that only 5% of the population is eligible, because you only have to meet the 95% cutoff once, on one of multiple acceptable tests. (And then there's parent recommendation, a whole 'nother thing.)
Second, the pool of "who actually takes the test" is further skewed. I'd guess at least 5% of my kid's school would be eligible to test, but she's the only one who actually did. Some schools sent most of their GT-identified kids. So not only are the percentile reports by grade not based on "all kids," they aren't based on "all kids eligible for talent search," either.
Third, there's no national reporting of talent search scores. Each search reports its own scores, and some of the searches are a lot more competitive than others.
For NUMATS, a 9th grader with a 21 Reading ACT would be 27th percentile, SAT-CR of 510 would be 33rd percentile, SAT-M of 530 would be 32nd percentile. A 9th grader with a 27 Reading ACT would be 65th percentile, SAT-CR of 610 would be 75th percentile, SAT-M of 680 would be 91st percentile, SAT combined of 1240 would be 77th percentile. (In the past, I've been able to find at least one other search's scores online, but no luck today!)
ACT Reading of 24 is 47th percentile; SAT-CR of 620 is 78th percentile; SAT-M of 570 is 48th percentile. (Notice there's also a large percentile discrepancy between the reading scores on the ACT and the SAT. I'd lay that on "those tests don't test the identical thing.")