It's about oversampling. Due how the survey is sent, a massive influx of machines coming online all at once will be more likely to trigger the survey. They know the general composition of their users, so they need the survey to be around the ballpark of that.
They are still only reporting the data they see. They are not correcting or manipulating the data like phoronix implies in their article.