Would you buy a GPU that you could safely run Furmark on 24/7 if it was limited to, half or 3/4 of the clocks? That's basically what you're proposing. I seriously doubt that you would actually lay money down for that shit. The market as a whole certainly wouldn't, they would buy the card that's faster under actual real-world workloads and/or didn't require manual overclocking to achieve its potential in real-world workloads.
The good news is that you can have it both ways. Modern power management is smart enough to figure out when you're running a power virus and it will downclock to those safe levels, while us normies can get competitive clocks when running our real-world code. It works pretty well most of the time.
Given that modern GPUs have complex frequency scaling algorithms that take temperature and power usage into account when setting clock speed, there's a good chance that a modern GPU will do exactly that.
> Modern power management is smart enough to figure out when you're running a power virus and it will downclock to those safe levels, while us normies can get competitive clocks when running our real-world code.
Which is perfectly fine, as long as the clock speed behavior is disclosed (which it is).
What isn't fine is a product running at such a high speed from the factory that running certain workloads damages it.