There is a deep deep bias on LW of thinking that truth is the only aspect of belief that has value.

It's not a bias, but an expected-value calculation. Most falsehoods are utterly useless to believe, along the lines of "The moon is made of green cheese." Merely affecting a belief-in-belief can be useful for the vast majority of other cases without actually spoiling your own reasoning abilities by swallowing a poison pill of deliberate falsehood in the name of utility.

The issue is that without possessing complete information about your environment, you can't actually tell, a priori, which false beliefs are harmless and which ones will lose and lose badly.

When you have a sophisticated meta-level argument for object-level wrongness, you're losing.