This issue could be solved over night if the USFG (NSF/NIH/DoD) stepped in and said "all publications supported by our grants must be published open access and we'll pay no more than $N/page in publishing fees."
You probably need to explicitly ban publishing in publications with publishing fees, otherwise money from other sources will be used to pad out the difference.
No, that's letting the perfect be the enemy of good enough.
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".