But of course, this fails completely under sybil attack (I know you said not to mention it, but unfortunately we cannot talk about consensus without it). If I own a majority of validating nodes I can do anything I like: come to a majority consensus on transaction A, then double spend with transaction B by dropping all record of A from my nodes and coming to a 2nd consensus. Or I can completely freeze the chain by censoring all transactions, etc.
Thats a moot point, if you own the majority of nodes in any system you can do whatever you like.
Now lets talk about the cost of this attack - in instant-x and zerotime, this cost is a simple constant in the amount of stake you own. In emunie, afair, this cost is completely zero, because of the long con attack.
I fail to see how you came to the conclusion that it is zero, when I posted upthread that the primary determiner of vote eligibility was now the most recent endorsements you had received, and not the overall long term amount.
Therefore if you want to vote on the outcome of a conflict, you MUST of done work and been endorsed recently. Historic weight with this change is no longer a factor, so the long term con as you put it is gone.
If you want to Sybil your way to eligibility, then you must constantly be acting as A, C and D and creating the majority of transaction in the network. As soon as you stop, you lose any advantage you had.