It's somewhat disappointing imo to see the Fury X's massive 8.6 TFLOPS (which I believe a heavily OCed 980 Ti can reach, not sure, but at that point the 980 Ti is pulling way more power than Fury X) not really translating to much in games.
It's always been a matter of utilization for AMD. NV cards have fewer SPs, but they tend to be more powerful, or in the case of Pascal capable of operating at vastly higher clock rates.
Fury has more theoretical compete, but it's more difficult to completely saturate all of its SPs. There was previous discussion on AMD implementing technology (based on a patent) to dynamically increase clock rates for a CU when not all of the resources were being used, but no one was sure if this was something that would make its way into Polaris.
DX12 also helps alleviate some of these problems as extra SPs can be used for async compute purposes which should at least keep the most of the SPs fed, but I think we're still over a year off from seeing games that are designed from the ground up for DX12 to take the most advantage of that hardware.
Hopefully this doesn't leave Nvidia in the dust if there will be game engines that are heavily skewed towards AMDs compute advantage.
It won't. By the time DX12 becomes mainstream, Volta will be ready. If they haven't managed to create an architecture to harness DX12 to its full potential by then that's their problem, but they've got too many talented engineers to seriously consider that as likely without some strong evidence. It probably just means that Pascal is not going to age well, but that's a different matter.
Game engines will probably skew AMD anyways though (unless NV pays a company to use Gameworks) as a lot of PC games are console ports and the consoles will all be using some form of GCN-based graphics so AMD gets a free boost from that alone.