If you take, e.g. two unranked players that are very bad at chess or two unranked masters they are still going to end up with a +16 and -16 elo change. And you can keep adding great or bad players to the pool, and elo is still going to only look at the outcome not how players play.

The point with elo is that if there's only 2 players it is basically impossible to reach a 2000 elo, because even if one consistently wins at some point he's not gaining any point by beating the same opponent, thus to go from 1700 to 1800 he'll need to face an opponent that has a similar elo.

The more people slightly below his skill the more he'll rise in the elo distribution, this trickles down all the way up and down.

Of course it is very likely that modern 2500 players are better as they have better tools than players of 20 years ago, but the same applies to people lower and higher in rating.

Thus, at the end of the day, the only factor that matters in an elo distribution and how wide it is, is the number of games and players.

If tomorrow there will be an influx of another million ranked players the distribution will get a bit wider and this would also inflate in the long run ratings of the highest rated players.