What is funny is the claim JPM make about Apple finding it so easy to increase A8 GB3 performance that much that quickly. They will only be able to do it mainly by clockspeed in single-thread as Apple's ex-Dec guys have already played their Alpha EV6/7 card and I don't see a mobile Cyclone derivative at 2.6 GHz even at 16FF. Cyclone also has trouble repeating its GB3 dominance in common application based benchmarks too questioning again the importance anyone should place on GB3 numbers. The fact that JPM does so much just staggers belief and cries out for the Spec org to release SpecCPU2000 as freeware to end all this bench-marketing nonsense.
Yep, first of all Geekbench will likely fit into A7s Level 3 cache and that doesn't represent the performance when the whole OS is running in the background with all those context switches and related cache misses - that's where a very complicated and mature cache controller comes in - let's see how Apple does there compared with Intel. AMD tried to beat that for more than a decade and they failed big time.
In addition, how much is TSMC 20 nm to reduce on power consumption? Not much if I remember correctly. Now, A7 is already a power hog with up to 10 Watt burst power. If they increase the clock frequency, they have to increase the voltage, which will cause more power penalty than it improves performance (in addition, the slower memory causes more processor stalls). If they double the cores they'll have a useless setup for mobile applications and they'll have twice the power they need at maximum load, which may result at 20nm to about 15Watt I guess. Going wider is only increasing performance minimally but power will go up too. At 15 Watt, you'll get a very powerful Broadwell processor. Haswell goes down as low as 5 Watt already. And that's true desktop performance with high SpecInt benchmarks.
Nevertheless, Intel needs to speed up its Core line significantly. They were too concentrated on bringing power down, which was important, but performance needs to increase more than 10% per generation.
The other thing is that Intel should release mobile SoC built on Broadwell. Especially for tablets this would kill everything on the market performance wise. Even phablets should be possible with such an SoC. Intel should aim higher once and for all.
Market Data 
Markets 
AI
