I agree that lowering curation rewards by 75% without increasing other forms of inflation would decrease the incentives to stake HIVE. As a result, the vesting fund would probably shrink substantially. Maybe not a full 75% as merely holding HP has a roughly two percent APR and there is some benefit in partaking governance and because of the possibility of vote farming.
Because author rewards would not be reduced, a new equilibrium would be found where the smaller total amount of vests would be able to earn the same curation rewards per 1000 HP as before.
Problem is that the remaining stake could be used for auto-voting just as well. See, the main reason for auto-voting is not the fact that curation is a lot of work - it isn't if you vote mainly on a reasonable number of authors you follow - but the fact that curation yields depend so strongly on the timing and the order of votes on a post. What's broken here is competitive curation because it leads to curation sniping, not curation by itself.
At the new equilibrium, the division of author rewards vs. curation rewards would be 80/20, which would be nearly the same as before the EIP. You can count on vote farming in the form of bid bots, circle jerks or straight up placeholder posts being self-upvoted or sock puppet votes to re-emerge. You think this would be kept in check with downvotes? I don't think so. Small stakeholders' downvotes wouldn't make a difference. The risk of retaliatory downvotes would be too great for the middle class. Could the whale population be trusted to keep the vote farmers in check? The problem is there are too few large stakeholders. They wouldn't have time to do that and in some cases they would be the ones doing the farming themselves. Quality control hasn't been working too well when it comes to the proposal system, either.
Also, I don't think automatic voting is bad per se. A lot of large stakeholders vote on a number of trusted quality authors who realize that starting to post shit to milk the autos would eventually cost them the large auto-votes.
What is actually problematic is content agnostic voting, automatic or not, to snipe for maximal curation rewards. What I propose instead is changing the parameters of competitive curation to make the timing and order of votes to matter less.
There are many great manual curation projects such as @curangel, @curie and @ocdb among others that pay for delegations. With the advantage in curation sniping greatly reduced, the curation rewards afforded by those volunteer-run curation projects could pay delegators substantially better.