After being on the road all day, I accidentally posted a comment that made sense. Let me try to develop it further here, so my bots will vote for it. It should at least make you think.
I wasn't planning to change Steem's reward system at all, but I responded to a post by @cervantes, who polled the witnesses about a number of reforms. Among other proposals, about half of the witnesses are conspiring to raise the curation rewards from 25% to 50%. I think this would only change the nature of voting abuse. But we need to go deeper. There's something fundamentally wrong with Steem's voting incentives.
Sorry @dan, the whole concept of curation rewards is ass-backwards. I always loved explaining to outsiders that voting on Steem doesn't cost you any money, and you're even rewarded for it. But it was too good to be true. There's no free lunch. Voting can be abused because it's free.
Curating is not a selfless act of sacrifice for the greater good that should be rewarded. It's an act of exercising cultural power. I vote for @traf because I want to see funny gifs. You vote for @trafalgar because you want to read boring essays. It makes me feel good to see the number under the post go up. Traditionally, this power cost a lot of money IRL, and it was one of the things that made the grandchildren of self-made men squander their inherited fortune.
So here's my proposal: voting should not be rewarded at all. On the contrary, it should cost money - more money than the author receives. For example, if the author reward is 100%, the curation reward should be -150%. Nobody would self-vote in this system, not even between different accounts. Voting bots would be unprofitable. Automatic voting would be reckless.

You're not getting a cookie!
There would be no curation window, no reward pool, and no inflation but deflation instead. Wouldn't that be good for the price? And if you ask me, we can do away with Vests, Savings and SBD too. If we want mass adoption, we need a system that the average person can understand immediately.
Would nobody vote in this system? Well, a vote would become much more meaningful. And we can still keep a system where the early voters of the best posts are rewarded in the end.
But we can't just assume that the posts with the highest rewards are the best. That would still enable spammers to curate the shit out of one of their posts.
If we really care about quality, we need a new class of witnesses: users who judge the quality of posts by their own voting behavior. These content witnesses would be elected like the block-producing witnesses, with one difference: everybody also gets a number of downvotes in this election. That way, we can keep out users who abuse the system - while downvotes for individual posts become unnecessary. Let's call these people quality oracles. They'll have the privilege of voting for free. There could be more than 20 full-time Quality Oracles, maybe 100. We'd need to find a balance between keeping out spammers and allowing a diversity of opinions and backgrounds. Maybe we'd even need quotas for other languages than English.
This proposal contains a lot of moving parts. I'd just posted another comment asking to implement only one change at a time, so we could determine which change actually worked. But in these debates I see a lot of assumptions that are taken for granted and very little research. We need to rethink everything in order to find a system that will work better. I hope that this proposal will help work out a system that takes human nature into account.