As others have said, ARM SoCs can't improve 2x every year forever. ARM has run out of the low hanging fruit with the latest generation of CPUs where 3-issue out-of-order execution architectures were introduced with Krait, Swift and A15 cores. For a period of 4 or 5 years or so, ARM went from simple 1-issue, in-order, non-superscalar architectures to modern techniques (superscalar, wider issue widths, branch prediction, out-of-order, SMP) on a yearly cycle. They benefitted from all the hardwork in the PC, workstation, and server space and was able to introduce these techniques over a 4 or 5 year time frame instead of the 2 decades it originally took.
Arguably, this was because they needed to fit in a 1 to 4 W TDP envelope and they needed to wait 65 nm nodes or better to even do it at those power consumption levels.
1. The only metric that is really important to most everyone is single threaded (single core) performance and the whole memory train supporting it. A 30% performance improvement over a years time frame is actually quite normal, if better than it should. Maybe Apple can squeeze out another 100% improvement, but it's getting harder and harder to do.
2. 4-cores isn't important to most everyone. Most software can't actually take advantage of that many cores. Software that can take advantage of it, anything involving matrix math or easily split apart tasks, are few, and most aren't fit for mobile handhelds. Transcoding video is not something done on your phone. So, dual-core is typically the optimum number of cores needed for 90% of the folks. What's really really important is single-threaded performance. Maybe in a couple of years, video transcoding and editing will be common enough or maybe there be some killer app that can make use of it, but definitely not today.
3. When talking about 64-bit, what people are really talking about is the instruction set architecture and addressable memory. It does not refer to data widths, like in a 64-bit memory bus or 128-bit SIMD/vector instruction. It doesn't mean it'll be automatically faster. Typically, what it really means is greater than 4 GB of RAM support.