So Elo is specifically constructed to be normal. Like, that's the assumption, and so saying Elo results in a normal curve is a little off because well, that's what it was designed to do. In this case, if it was standard normal then rank 95 would be where the top 5% land, but it's clearly much, much steeper. It's possible that they are using a conversion or a different way of handling rank separate from the PDF, but it would be odd to use a similar, but not exactly right, set of numbers.
Another interstitial thing about Elo: the good professor made the assumption that everyone's variance would be equal to make the math work, but he suspected that wouldn't be the case. Indeed, it is not; lower skill players have a significantly greater degree of variance than higher skilled players. Modern improvements to Elo have tried some different stuff, including the logistic distribution, which might be what's going on here, but it's hard to tell. It's just so odd the tails are so long and there's so much movement in the middle.
I'd really be interested in finding out what's under the hood of assorted competitive game rankings, but they don't often talk about it publicly.
