As far as I remember, the main problem with emulating Windows on PPC hardware was the fact that PPC was big-endian and Intel was little-endian (thus, each CPU instruction had to be translated (inverted) by the emulator, so you couldn't get half of the host's CPU speed in the Windows VM (at least twice instructions needed for one request)).
I'm not an expert in deep CPU understanding; I'm wondering how similar would ARM vs Intel be against PPC vs Intel (as far as I know, both Intel and ARM are little-endian, so it's not the same problem as PPC vs Intel).
As far as I know, ARM processors are both big-endian and RISC based (X86 processors are CISC based). This means that, besides loosing time adjusting the endian for each instruction, the ARM processor could need to use multiple RISC instructions to handle one CISC instruction (as you, I am not expert on deep CPU design, so I hope I am not saying anything too stupid LOL). Result: SLOOOOOOW.