If a group is 5% of a population, then a meaningful sampling of that population will show that group as 5% of the polled people. Simple as that. Even where other demographical information comes into play, you'll see those lines show up in the sampling. Saying, if Linux was 50% of one region and 1% the others, the sampling would (accurately) show the 50% of users sampled in that region, 1% elsewhere, and the X% average over the entire population. That's the first core piece to understanding statistics and even political science. You not agreeing or disbelieving it does not change how mathematical sampling works.
This kind of sampling accurately finds far far more minute demographics in far smaller population pools than you have with Linux users among Steam gamers, so it's pretty silly to think that Linux users are so magical and special that math doesn't apply to them.
It's accurate to question _why_ the numbers are the way they are and question certain conclusions, maybe claim that there are more Linux gamers who simply don't fall into the demographics sampled by Steam (which exclude Linux gamers that reject Steam or many Linux users who game on Windows due to the lackluster Linux game library size), but to claim that the sampling approach is invalid is just wrong.