Most milk bottles I've seen list volumes in both litres and pints, which solves the labelling problem fairly well in practice. You can usually work the difference between 1/2/4/6 pint bottles out from shape and size alone. I've seen 2 litre bottles in the past, but they're pretty rare.
The butter I have to hand is in 250g blocks, so is entirely metric-based for now.
I'd argue that this all comes down to patents. The OoO/speculative execution methods in the P6 architecture would be patented, and those P6 patents ought to be expiring about now. That means that using the fast Intel method of speculation would make sense if you are designing a brand new processor with OoO execution. The older ARM chips did little/no out-of-order or speculative execution, so the cost of adding any particular variant is probably similar.
Meanwhile, because AMD was competing with Intel directly on x86, they had to produce processors with OoO execution for years to keep up. The P6 patents meant AMD created their own implementation of OoO execution, and now the patents are no longer an issue, the cost of redesigning the chips meant that AMD stuck to their existing methods.
There's a reasonable argument for moving to 64-bit on security grounds too. The increase in virtual address space makes ASLR far more effective since there are many more options for positioning compared to 32-bit code. On top of that, any attacks are more likely to hit a unallocated page as opposed to anything useful (with some limitations of course).
You can do that easily enough as it is. One way is to set up multiple SSIDs per radio with separate PSKs. Another way is to use WPA2 Enterprise with one username/password pair per device.
Save energy: Drive a smaller shell.
