If the vulnerability is part of a feature designed for niche use cases, then it's far safer than one that affects general use. I highly doubt most people are going to run virtual machines, plus the main target is server hosts that use VMs to run multiple servers of the same type on the same box. I might run a VM at some point in the future, but when I do I'll take steps to avoid any issues, like only enabling virtualization in the first place when I need it. Sure, that means I need to boot into the UEFI before and after every time I run a VM, but that's not an issue on the system I'd be running it on. And I'd rather have that inconvenience than have to worry about a vulnerability at all times.

In short, it's a matter of risk management.

At this point, I almost feel like we need to start over with the idea of pipelining in CPUs, as though it were some kind of original sin. The fact that the most basic of errors in pipelined logic are referred to as "hazards" should have been a hint.

(Edit: only half kidding)