Convergent linear should be very close to a 50-60% tax at the lowest end
From the graphs, it was like -40% value right at the start, and +20% at the peak
I don't know if detecting low level profitable farms is as you as you say. Keeping in mind many downvotes are in the hands of unsophisticated people. Put it this way, imagine Steem was $15 and there's ample incentive to try as hard as you possibly can in sophisticated ways to vote farm while avoiding detect. Making multiple seed accounts, then spamming RC to create more accounts from those seed accounts, then spreading them into clusters with only minor overlap, but moving funds and votes in a way that's dynamic, with comments being a combination of AI and generic messages in a foreign language etc.
Basically imagine you tried your hardest, or even someone considerably more technically sophisticated tried their hardest to evade detection under linear, and compare it to under convergent linear. I'd bet the former would be far more elusive.
Remember, n^2 was someone with 1000x your stake got a vote 1,000,000 your weight. Compared to that, this is nothing.
Also, I believe over time, as your curation project gathers more momentum, it'll easily break out of the starting part of the curve.
RE: The EIP and how curation will matter again - a.k.a false hopes