The challenge with rewarding downvotes is partially in the fact that it is so easy to reward upvotes. If a piece of content gets paid a certain amount, then part of those rewards are shared with the curators. In essence, it is a profit sharing model the rewards earlier upvotes that theoretically took more risk on upvoting than later voters. It is easy to quantify if the upvote was worth it based on the resulting payout.

How do you quantify the success of a downvote? If it made sense to concentrate downvotes on bad content, then you could reward those downvotes in a similar fashion to upvotes. But then you run in to the awkward question of how you would reward an upvote on content that made 0 STEEM. It doesn't make sense that you would reward the upvote. After all, the community determined the content was worthless. Why would you reward someone for thinking the content was valuable? That situation highlights the intuition we have when the rules are mirrored that somehow get lost when we look at a downvote in a sea of upvotes. We are open to suggestions on how downvotes might be rewarded and agree that it would be the ideal solution. Our goal with the EIP is not to nail down the ideal, but simply and carefully move closer to it, one step at a time.

Regarding second price auctions, the idea is interesting, but sadly there are simple behaviors that will get around it. If the highest vote gets thrown out, then I will split my stake among more accounts and continue self voting. Then only a fraction of my stake would get ignored via the second price auction rules. Your simulated results are better than expected because they don't account from any emergent behavior as a result of the change. This is one of the biggest challenges faced by the scientific community with regards to social sciences. I appreciate that you are thinking through these problems as well and trying to come up with solutions!