The only reason I'm not touting translation/emulation as the cure is that, AFAIK, the only current "proof of concept" is the x86 emulator on Windows 10 for ARM which is 32-bit only, which would be as much use as a chocolate teapot on a Mac.
I'm not sure if that is for technical reasons, if the CPUs that MS is using aren't fast enough (maybe the A-series processors are better) or if it is legal blight (Intel were even being sniffy about x86-32 emulation at one point and since x86-64 was originally developed by AMD, the licensing for that is probably an even bigger can of worms...)
But, yes, if Apple can come up trumps with a x86-64-on-ARM64 equivalent to Rosetta, that could be some help. However, this time around, they may even be able to do without it - the computing world has changed since 2005, a lot of modern Mac apps - especially those aimed at the low-end Macs - really are going to be 'check the ARM64 box and hit build'.
In 2005, a Mac - even a low-end, consumer-focussed one - that couldn't run Office or basic Photoshop would have been a dead duck. In 2020, more and more people in that category are using platform-independent web apps for basic productivity and there are a whole new generation of "prosumer" graphics apps which are already available for iPad (even Photoshop itself - although I understand that still has a way to go, it's better than starting from scratch).
If Apple repeat the PPC to Intel process and release a developer's machine in June, there could be a metric shedload of "native" ARM Apps by the time consumer machines launch next year.
Translation won't help so much with the big pro application stacks - the only cure for that is a decent transition period.