I agree that there is a centralizing effect when large miners mine spam, and I blame that centralizing effect on the people who create spam and the miners who mine it, rather than on the people who filter it. I also think relaying spam has a centralizing effect because it disincentivizes running a node, and I think the mempool policy adopted in Bitcoin Core is directly responsible for part of that centralizing effect. By contrast, Knots is not responsible for the centralizing effect produced by golks who bypass Knots's filters. Thus, both approaches involve a centralizing effect, and an important question is, which effect is worse? One is directly attributable to the mempool policy in Bitcoin Core, which welcomes and redistributed spam; the other is directly attributable to spammers and spam-miners who build tools to bypass Knots' filters. I think the former is worse than the latter.