impossible to say. If AMD goes balls out, they could theoretically make a 8192 core monster GPU on this node if they wanted to.
AMD doesn't need to do that since CF scaling is not 100% across all games. If CF scaling was consistently at 90-100%, you'd have a point. Just 25% faster than Fury X is going to provide similar performance to max OCed 290 in CF, but in all games, regardless of CF scaling.
Even now a max overclocked 980Ti provides better gaming performance than dual 290s in CF. It's a foregone conclusion that when comparing 290s OC CF, Vega OC will beat them in smoothness.
Or sell the 290's and get TWO VEGAS! But yeah I bet a single Vega would be better in most cases. We'll know in two months. Two loooong months.
Vega is launching late 2016/early 2017, or probably 9-10 months away.
Assuming we find out much about Vega at Computex.
I don't know why anyone would get two Vegas though. SKL-E is going to support three full x16 slots, so it would seem like V10 Trifire should be a standard system configuration come 1H2017.
PCIe lanes don't matter anyway, at least for AMD. Even going all the way down to PCIe 1.1 x8, the performance hit is only 6%. Fury X Tri-Fire on PCIe 3.0 x8 on a 5820K for a 1% performance hit at 4K. Who cares.
OP, set up Ethereum mining on those rigs and you'll get free Vega CF/Tri-Fire upgrade. Nothing to think or worry about since you already have the cards and they are quiet courtesy of water-cooling. Every day you miss is
$7 lost in earnings. In about 95-100 days, that's already 1 Vega card paid for and that's not counting the resale value on those 290 cards towards the 2nd Vega card.