Honestly, that doesn’t seem very bad, considering the variability of “going feerates” and the practical realities of block confirmations. The “core” pinning problem is a 500x mismatch in pacakge sizes, not 25%.
anti-DoS rule#3 is the pain point here. If it gets evicted from the mempool, then you’re in the clear(future bytes will have to pay higher minfee, protecting the node), otherwise you have to overcome the total value.
The future post-cluster mempool scheme I have in mind is pretty simple: v3 would be relaxed to on the order of “must be in top block to enter mempool”. This makes pinning essentially impossible unless the attacker somehow knows the top-rate backlog is going to get bad shortly in the future, and should protect against witness inflation and the like. Would allow for pin-resistant batch CPFP, would mean you don’t necessarily need no unconfirmed ancestors, should make ANYONECANPAY usage more safe(like HTLC second stage txns) etc.
Just a sketch at this point, but something to consider if you’re already thinking about pins and mitigations.