And don't get me wrong there is a point to have benchmark that tests all these bursts and stuff for certain usage models this makes sense. All those fancy instructions with vectors and such is not important if you are not doing anything that takes advantage of vectors and things. "Normal people" (I am using this as a pejorative) may have no need whatsoever for those instruction sets, but those instruction sets provide an order of magnitude improvement.
But then again you use a can use a mobile cellphone differently than a tablet differently than a light and portable laptop different than a desktop / workstation. I would not wanting to be doing lots of data entry on any cellphone and if I did I would want a blackberry with an actual keyboard. Same thing when we start moving to more complicated stuff such as engineering, graphics, video rendering, scientific/math calculations, program compiling etc.
Like I said earlier Geekbench is useless to me. It is just not good, it may be better than other mobile benchmarks but it is still not good or useful.
Also your a8 about throttling is also useless for me for unless you are playing a video game 1.2 ghz vs 1.4 is plenty of speed for a cellphone (with apples a7 and a8 ipc). I want burst speed and dynamic range for a cell phone usage model, you know race to sleep and quick to ramp when I want it. 4 minutes before throttling is plenty of enough time. And if you are playing a video game setting the cpu goal to something like 1 ghz instead of 1.4 makes sense to keep thermals down, maximize battery life, allow n-1 or n-2 generations to play the game etc. That said throttling to 50% cpu speed instead of 75% cpu speed looks really bad (5s numbers, 6+ is good in my opinion)