I vaguely recollect from reading the blog Language Hat that modern linguists sometimes decide which variant of a class (for instance, 'software' vs 'a software' vs 'softwares') is more correct by counting how many Google search results each variant returns. No doubt they need to filter this data in many ways, but it would seem that it is no longer possible to say 'X is correct, all else is not.'
A better example might be people using if when they should be using whether.
First, I think there are degrees of nativeness in language acquisition. Cutting out the entire class of strictly non-native speakers is a bit of an extreme step.
Second, what about differences in usage of the same language over two classes of native speaker?
http://en.wikipedia.org/wiki/African_American_Vernacular_Eng... was once considered to be "incorrect" English.
Unless you're the Mad Hatter, you'll be ordering a (whole) glass and enough water to fill it. We collapse the "glass of" into epsilon because it's assumed in polite conversation the waiter won't be pouring the water into your open hands.