With iOS you have a real argument, in that the advantages that come from being able to move the platform along more quickly and deploy resources much more efficiently weigh directly against loss of old software that is no longer maintained (games in particular, a lot of real gems didn't make it from 32-bit).

With macOS however, and for that matter Windows or Linux or anything else on the PC, I think it's arguable that the last 10-15 years of improvements and universality of virtualization (and emulation for that matter) have changed the calculus completely. Even with Windows you say

>Windows software doesn't work 100% of the time

But in my experience it does work 100% of the time... in my Win98 or Win2000 or WinXP VMs which are trivial to spin up at will. And I also have macOS (previous Mac OS X or OS X) VMs going back to 10.6 Server, the first support virtualizable version. macOS is more highly dependent on hardware acceleration, but even there the advent of PCI passthrough means it can be solved at negligible performance penalty. And for OS that are even older, there is now such a delta in raw CPU power that emulation can cover it.

So I think the balance has shifted in terms of the value of maintaining BC in a PC OS. It's not useless by any means, nobody wants software breaking every single year. But compared to even 5 years ago I now lean towards it being fine to start leaning more on virtualization beyond the 5-7 year mark, and certainly the 8-10 year mark, whenever there are real benefits to moving forward.