64 bit x86 is not slower
Because 32 bit x86 is so register starved, 64 bit x86 is faster in most cases. 32 bit ARM isn't register starved, but 64 bit AArch cleaned up the API and added some new stuff like CMOV instructions so it too may be faster in the majority of cases.
When Apple shipped the first 64 bit ARM SoC three years ago, the results of benchmarks run on it (as well as Apple's internal results) show it is faster running 64 bit code because of those API changes. Had it been slower, there probably would have been less of a rush for everyone else to catch up, since 64 bits is otherwise useless until you go beyond 4GB RAM in the device - something which is only now happening with a couple Android phones shipping with 6 GB.
If you have code that is very pointer heavy then doubling the size of the pointers shrinks the effective size of the cache and can make it run slower, but this is a pretty small effect in the single digits. If you have code for which this is a problem you can continue running the 32 bit version even on a 64 bit OS, at least until 32 bit support is ripped out of the CPU entirely (Apple will probably do that with the next major redesign of their 64 bit core, but I doubt anyone else will for a long time)
Biting the hand that feeds IT
