[Sanddancer]

ARM's 64 bit is pretty much always [1] dragging just as much legacy baggage as Intel's 64 bit. However, that's fairly irrelevant once you get everything decoded into micro-ops and sent into the pipeline. There, branch prediction, pipeline depth, number of pipelines, etc all matter a lot more than the opcodes the CPU sees.

Also, you have to keep in mind that Gruber's been banging the single core benchmark drum pretty heavily, and rather loudly ignoring that Apple's ARM chips are very inefficient in multiple core benchmarks. In Geekbench's tests, Intel chips scale at nearly 1:1 with additional cores [2], while dual core performance of the ipad is only 70% better [3]. So, any sort of operations where you're wanting multiple intensive threads going on, the A9X falls on its face. So when you start looking at the hardware in the light of a heavier workload, Apple's not gaining that much ground on Intel, who has a lot more experience going broad than Apple does.

[1] AArch32 is optional, but pretty much on every ARM chip for compatibility

[2] http://browser.primatelabs.com/geekbench3/search?q=6600u

[3] http://browser.primatelabs.com/geekbench3/search?q=iPad6%2C8...