EDIT: To clarify, it doesn't have to be that stupidity causes bad grammar, but if bad spellers are likely to be bad commenters, that's all you need.
Establishing correlation without causation is a minefield, IMO, although each of us may come to different conclusions. Filtering people on the basis of correlation and not the primary attribute is discrimination. I doubt filtering a forum is illegal discrimination, but I’m uneasy about where that might lead me personally if it were my forum.
Giving a ridiculous example, at a certain time in history, correlation might have established that 90% of the people who used AOL to get onto the Internet posted idiotic things. Should all AOL users have been banned from forums? I suspect that many forums would have had a net positive benefit from such a policy, however the 10% false negatives disturb me greatly.
I would morally prefer other mechanisms for filtering the 90% out, preferably mechanisms that directly test user’s propensity to the “post idiotic things.” This is a personal view, and I can accept that others may not agree.
p.s. And then there’s the question of gathering actual data and not anecdotes. The worst way to proceed would be to claim that since we seem to remember that 90% of all idiotic comments contained grammar mistakes, there’s a 90% chance that a comment containing a grammar mistake is idiotic.
Those who miss nuances of thought and logic are often the same folks who miss nuances of grammar. It's a skill one can learn. Using that as a membership filter would be no different than forming a programming group and restricting it to programmers.
Your argument reminds me of the argument that someone ought to wear a suit and tie to work: There is no excuse for not knowing that a suit and tie is what people wear to work, and anyone who doesn’t do so is not fit to work here.