So, by "meta level" I assume Will just means "the rest of Plantinga's argument" which is that if beliefs aren't reliable than there is no reason to believe naturalism (not a reliable belief).

The key of course is the phrase "perfectly reliable" which Gwern uses and Plantinga does not. Plantinga admits that beliefs are not perfectly reliable- when he says reliable he means something like "the vast majority of the time". He has a specific argument that a belief causing adaptive behavior is not an indicator of its truth. It goes something like this: behavior is caused by combinations of beliefs and desires. But many (most?) beliefs and belief-desire combinations that would cause adaptive behavior are false. For instance, a hominid would behave adaptively if he desired to get eaten by a lion but falsely believed that any lion he saw would try to protect him from other lions. Similarly, there is nothing to keep evolution from selecting from beliefs that are false but don't impact survival behavior (like, say believing trees are trees vs. believing they are witch trees.

This argument is weak: evolution acts not on the set of possible beliefs but on the set of available beliefs. Plantinga has to argue that evolving the false-but-adaptive beliefs is just as likely as evolving true beliefs. Opponents of Plantinga need to develop this argument into something more robust and explain specifically why beliefs about practical things like simple observations are reliable. (Having a single method for forming new beliefs about evolutionary unprecedented threats seems a) more parsimonious and evolutionarily available and b)likely to produce reliable beliefs.

That there are cases of routinely false belief that are ruled false only by reference to beliefs we have reason to trust is just icing on the cake: theism doesn't have a good explanation for the existence of cognitive biases if it is being using to explain reliableness. Don't bite the whole bullet; just half of it, catching it with your teeth.

There is also the matter of methodological naturalism vs. metaphysical naturalism. The former renders Plantinga's argument useless and is a more reasonable position than the latter anyway.