There's also a question of where they're measuring this.

You can't intercept and inspect the browser telemetry of every web request on the entire internet, that's just not possible. So they must only have access to the data from requests from a specific set of websites.

Are those websites a representative sample of all web traffic, or are some of those websites more or less likely to have Linux users than the average website?

And it seems to be counting per request. Do Linux users visit these websites more often or less often than average? Do they spend more time there and view more pages, or less time and less pages?

A truly unbiased OS survey is basically impossible. At least, hopefully, they're using the same measurement data over and over again through time, so maybe the trends in the data are still meaningful.