This miner has some cpu load issues, nicehash team noticed about it in there first posts.
Miner involves cpu in removing fake results in last round.
The duped() function is slow, especially on celeron, in 6-card rig GPU load is only 92-93%. The host code must be rewrited, allowing executing GPU code and CPU code simultaneously .
The second issue is speed reporting, it causes some cpu load.
I made some dirty fixes, GPU load is 97% and speed went from 395 to 435 on overclocked 1070, but it is slower then Ewbf's miner.