> for zero benefit

I dispute this. When most users censor spam in their mempools, it pushes up the cost of getting spam mined. I think that reduces incidence of spam to an appreciable degree. This is a benefit and it comes with additional benefits, like reducing the centralization pressure spoken of in my slippery-slope post ("Spam-filled blocks lead to...a network [that]...is centralized.")

I want to weigh my proposed upsides against your proposed downsides, not just ignore them. Do my upsides outweigh the centralization pressure you speak of? I think so. At least up til now, it seems like the centralization pressure created by private mempool usage so far is marginal. The extra income accrued by those miners is tiny compared to their "regular" income. But the costs imposed on spammers is relatively massive: their costs double or sometimes even triple.

Similar considerations apply to fee estimation: very few fee estimation tools "only" look at what's in your mempool, and so far no one has complained that they couldn't get their transaction mined. If it eventually gets to that point I would support removing the filters but I see no evidence that it is a serious risk.

I am not familiar with your point about compact blocks. Can you elaborate or link to somewhere this point has been discussed at greater length?