The high clocking on Pascal really looks like a we couldn't hit our performance targets, so crank the clockspeed. And the cards are stock clocked close to what they can handle, since boost clocks sorta suck once thermal saturation sets in. Time will tell if water\aftermarket coolers will solve that one though.
As for AMD's side, it keeps being pushed around that the cards are being aimed at mainstream, and thus acceptable performance at a low price point (read, lower power requirements so OEM's can skimp on cooling+power, and cards can be made cheaper) seems to be the gig they are pushing, so I'm loathe to try and compare Polaris 10 to Medium Pascal. However, I can't see anything stopping Polaris 10 cards clockling like Pascal cards and AIB partners from releasing Polaris 10 cards with higher power delivery, with the limiting factor likely to be since the die is smaller, it'll be harder to disperse heat compared to a 1070.
As for GCN upgrades compared to Maxwell\Pascal, I just don't think its quite comparable from a generational leap, as we know the root problem is Maxwell\Pascal is that they need to switch between graphics and compute loads (okay when your being fed DX11\OpenGL serial workloads), whilst AMD can execute both in parallel (which is great for DX12\Vulkan, and the whole reason that AMD gifted Vulkan to Khronos and that Microsoft scrambled to implement DX12, and which left Pascal as an reactionary architecture by Nvidia). The only thing huge IMO is predication of loads, reducing the switching overhead, and helps to minimise AMD's advantage in parallel workloads.