SLI in one big respect has been a thorn in nVidia's side for years. It made infinitekly more sense to buy 2 x70s than a single x80 because you'd get 40% more performance for the same cost. Scaling was 70% with 9xx for example in TPUs game test suite but the big AAA games were 96+%. Yes there were games with no scaling, even a slight negative sometimes but they were rare and it really didn't matter... so what if a game didn't scale when it was delivering 88 fps with just a single card, the x80 was delivering 96. So was it worth being 40% faster than the x80 on average when fps was in the 35 - 70 range because games up at 85+, no one botheresd to put in the effort to enable SLI ?
Then something big happened.... AMD had nothing to compete with from the 1060 on up. After suffering cannibalized x80 sales (where they make more margins then selling 2 x70s) , nVidia had an option... what if a pair of X70s did not beat the 1080 at same cost ... cost of x70s went up.... scaling went down ... just 18% at 1080p... 33% at 1440p ... but strangely enough, at 4k, scaling was > 50%. This essentially made the twin x70 case untenable and the x80 the smarter buy @ 1080 / 1440p. Unfortunately... the 1080 Ti is still too weak deliver the hi fps that gamers want today w/ 6 of the 20 games in TPUs test suite unable to achieve 60 fps.
And now with the imminent release of the new 100 nit, IPS, 144 Hz HDR 4k screens, I suspect we will see that 50% grow a bit as nVidia will want to be the card choice for those aching to step up to these new screens. Most folks have stayed away from 4k so far I think as they have become accustomed to playing at 144 - 165 HZ. The 15% @ 1080 p => 50 % at 4k spread makes hard to discount that there's something intentional here ..
As for cooling, I have 2 issues w/ hybrids... most vendors are pairing their hybrids with their weakest PCB. In addition, mosy hybrids are GPU only leaving the memory and VRMs to be the chokehold. If ya don't like the idea of adding a full cover block ,MSI's 1080 Ti Seahawk (w/ EK full cover water block) is the best option. For twin SLI, I recommend a Swiftech 35x2 Pump (dualie), in single loop feeding CPU water block, and splitting the lines into 2 parralell feeds to the GPU blocks then recombining. Because of the large mass of the blocks, you don't need a large flow rate ... Im seeing 39C on each GFX card and 75C on CPU at max OC under stress testing w/ 5 x 140mm of radiator.