Wouldn’t deleting OP_RETURN CAP unintentionally give users an incentive to choose the cheapest or most cost-deferred method?

Why is that happening? The limit does not make OP_RETURN more expensive, and if there is a limit it will make OP_RETURN cheaper, and the cost per byte of data is the same regardless of the limit. The increased limit is that OP_Return is more useful in situations where you want to add 80 bytes or more of data to the output. This is great for everyone as an output data substitute adds that data to the UTXO set.

Instead of treating all vectors equally, shouldn’t a policy try to manipulate the demand for any data for the “most harmful” output, like Op_return?

That’s what this policy actually does. It incentivizes people who store more than 80 bytes of data in the output. Otherwise, you’re using bare multisigs or multiple Taproot outputs to use good Op_return across your network.