It’s a question of probabilities and assumptions, there are no absolutes here. Every time a P2TR keyspend is done in lieu of a P2A spend, that’s ~46vB additionally used.
The actual % of pin relies on many factors, the absolute worst case being a very flat, deep mempool, a tiny commitment transaction and tiny honest CPFP.
To reverse the framing, every time you require a signature where none was needed, with small packages you are deterministically “pinning yourself” by ~10% to avoid being pinned by X% due to a completely unrelated third party attacker. As packages get larger, the relative pin in both senses drops, of course.
A mining pool may stand to profit nicely if the victim pays 50% more in fees than they normally would
I think this is isomorphic with miner censorship with one exception: the fact that there would also have “victim miners” who are essentially being robbed of the “honest” CPFP that would have been on the table initially for the next block. If miners in general think they’ll get more out of you by collectively waiting, that cannot be helped and you need to set longer HTLC expiries to make defection more palatable.
All that said, I think this specific choice is a fairly minor decision in the overall security story.