There are different ways you could implement it, but the way I did it still uses time of vote in determining rewards. Voters get credit for their own voting time with the next lower rshares, and it uses the same square root weighting as today. So early voters would still get the lion's share of the rewards. It probably wouldn't make sense to wait 'til the end.

You're right about the possibility of splitting one stake into multiple accounts. I mentioned that in the article. But, if you have a small number of accounts, you have to sacrifice a large percentage of rshares. If you have a large number of accounts, there's less of a built-in penalty, but it's easier for others to find it and downvote it. (and someone would have to decide if the largest downvote is also going to be ignored as an outlier. I did not include downvotes in the simulation.)