Tegra Next will be definitely more efficient. Like 10xx is vs. 9xx too. OK, it won't run in handheld mode at highest clocks, probably half of that or below, but it doesn't need to anyhow, as the resolution will be definitely lower, so the performance will be there.
10xx vs 9xx will be a bigger improvement because it's 16nm vs 28nm, whereas Next vs X1 is 16nm vs 20nm.
As NateDrake said, he talked about architecture. What it means is that clock isn't the only problem then. Core count will be and it might be the unknown component here, if it's a custom part and not Tegra Next.
As I said, if they end up using 1SMM, it means it'd be between 76 to 128Gflops if clocked between 300 to 500Mhz. That is a worst case scenario of course, but all I meant to say is it's not necessary a news implying processing power being higher than X1.
I still don't see the point in overclocking the dev kits if it's only 1 SM. I'd say that X1 is likely the minimum to expect, though it could be slower by a little bit. There's not need to have a dev kit over 2x as fast as the final hardware.
We dont know if the devkits were X1 overclocked. Just that it had a fan, like Jetson Tegra boards. Then again, 1SMM is just an exemple for my point, being that Pascal based doesn't indicate a processing power or an improvement over X1 on that matter. Could be worse, could be equal but with lower consumption, could be better at equal consumption.