http://www.gamersnexus.net/guides/2710-amd-relive-vs-nvidia-shadowplay-benchmarks
its hard to get a benchmark to say anything particularly is wrong or right, a lot depends on the user. At 1080p, even 1060/6 can have a hard time, depending on exactly what game is used, and the level of details etc. I have a gtx970, and on my hardest hitting game (MGS:V) at ultra settings with 4k DSR, my gpu (asus strix 970 with 24% user OC) is seeing 99% usage (i7-3770k at 56%). And that's an older game. Even with the newer games and the stronger 1060, its possible to hit 99% usage there too, and then trying to record with pure gpu hardware accelerated encoding, something will have to give.
Yeah, FRAPS was a bad decision on my part. Yes, software encoding using the cpu can have a bigger possible impact. There's a lot of different variables and i'll bet in 99% of the situations, you'd be perfectly correct that ShadowPlay would be the best idea for lowest performance hit, but that's mainly due to the extra overhead most users see with the more powerful gpus. With a 1070 on 1080p , its doubtful there'd be any measurable impact at all, but in my case, recording MGS:V, I'd probably be better off taking some of the load on the cpu and off the gpu as my cpu is more capable than the gpu in that situation.