Mempool policy, especially a mempool policy adopted by a minority of nodes, is ineffective at preventing other users with demand for transactions that do not adhere to the policy from being mined. This is demonstrated by the 20+ new OP_RETURN outputs in excess of the limit. As about 18% of the hashrate mine transactions with oversized OP_RETURN outputs, anyone who wants to send one will see it reliably mined if they manage to get it to one of the corresponding miners. They just cannot get it reliably in the next block.

This means that:

So, you don’t get the potential upsides of allowing larger OP_RETURN outputs, but you get additional mining centralization pressures from slower block propagation, and financial incentive to build out direct submission.

At this point, it seems better to make these transactions available to all miners, so at least the miners benefit fairly from the additionally available fees, and so that parties who pollute the UTXO set with unspendable outputs carrying data can at least be encouraged to use OP_RETURN instead as a less harmful option.

He thinks it's bad that spammers have to pay typically higher fees out of band. This seems to be the common argument given in favor of the PR.

In reality this is a direct result of the spam filter working. Spamming should be expensive as possible.

They take the desired outcome and twist that to be the reason why we should get rid of filters.