Outright bans make open access harder; editing+publishing with reasonable quality and archival levels of access guarantees can be cheap but it's never free.

Just limit it to a very reasonable $/page. Even upper bounding it at some obscene amount like $10/page would be a vast improvement and a completely trivial expense (you don't want to know what plane tickets to IJCAI cost this year...)

To clarify, I'm saying any research funded by the NSF should have this requirement imposed on all publications regardless of funding source. I.e., DON'T say "NSF $ can't be used for more than $X in publication fees", say "NSF $ can't be used AT ALL if you ever pay more than $X in publication fees".