You get same return no matter vote for what, the only factor is downvotes cast by others.
this is true.
The pure-linear approach work best in the situation where the community members are actively flagging those bad-quality posts. In a upvote-dominated system, where only a few minority doing policing job, then the flat reward-curve will just incentivize people to cast random votes.
so, linear-approach would work better if we could have a mechanism to compensate the down-voters. for e.g. maybe we could revise their voting power if a down-votes surpass some threshold (say, >50% of upvotes) so at least they wont be penalized on their voting power by doing good flagging job... it would be best if could come out with some rewarding scheme.
Good point. But keep in mind that downvotes can be abused as well, so hard to do the incentives right. As of now, I strongly tend to use an off-chain rewarding/compensating mechanism.