Some raw data for direct comparison (win 8.1 64bit)
970@1478MHz = 147MH/s (reported maximum 157), client-side

1070?
http://www.xtremehardware.com/images/stories/NIKOLAJ/Roundup_GTX_pt1/algos_new/ALGOS_2_mod.jpg1070 AMP Extreme @2052MHz = 268MH/s (tpruvot 1.78). On 1.8.1 it's no more than 278, I should recheck this value soon. With a multigpu setup, a 1070 with the latest 1.8.1 hashes at 260/even 265, with an average value of 255,258 /Rig (a VGA sometimes hashes 224MH/s with the other ones at 260,265).
If you would like to put some numbers with the same frequency, sp, it would be gladly appreciated.
My record in efficiency on a 1070 it's the following one:
81.5W, 202.77MH/s with the 1.7.8 miner

I get your point, but I would prefer to pay double the amount without fee (15% real, not 11.66% that translates in 8/9% stable with the maximum oc possible*), in all honesty, but a payment plus an high fee (3.33%) it's a lot.
On Yiimp alone you would gain 0.3BTC /day just in fees, 9 BTC/ month in LBRY (too many, you could be the one dumping this coin on a daily basis)
Thanks
* if you say 15% max, it's related to a single vga. Whoever use this miner in a rig with 6 models, will use more conservative settings, so it would be 11% more or less (stability). An optimization usually means worst W/MH ratios, so what you can achieve
under the 10/15% more, you could with the public version too (with a tiny inferior efficiency ratio). If you add the 3.33% fee, the net result would be 8/9-3.33 which is half, more or less, of what you, kindly, promoted.
It's just a deep analysis of the ratios, nothing personal.
I suggest you to release 2 versions:
1 with an higher BTC amount and no fee, and another public with a 2% fee, like all the other devs (claymore in primis)
SOURCE:
http://www.xtremehardware.com/recensioni/schede-video/roundup-gtx-1070-e-gtx-1080-analisi-e-confronto-dei-piu-recenti-modelli-custom-nvidia-pascal-parte-1-2016080211648/?start=22Some more data:
https://bitcointalk.org/index.php?topic=1586147.0