But they are into computers for computers sake, so to speak, rather than into computers as a tool for doing a different job.
Meaning they are looking for agnosticism, much like your server example, but on desktop/mobile/embedded.
And those seem to be what gets the hardest squeeze in all this, and are the least articulate in their complaints.
The irony is that many of those that are in that segment may know more about the day to day inner workings of things than the actual devs writing the code. This because they see the code in action with all the ragged edges exposed, rather than in pristine code form.
The key is to find what can be agreed on and where the split needs to be.
If my foggy memory serves me right, the windows embedded team took action on an analysis like this about 10-15 years ago.
With desktop/mobile/embedded applications, you often have to do a lot of hardware-level integration to provide the best customer experience. This requires lots of features to support different use cases like hibernation, cellular radio on/off, power control, realtime device support, drivers that support many different drivers on different platforms, etc.
With servers, you actually want the OS to be hardware agnostic. Tight hardware integration is a drawback - not a feature - since it makes it harder to migrate / clone a system. Performance is pretty much the primary driver on a server, so having fewer abstraction layers should improve speed. Simple is better, because there's usually enough complexity in the applications themselves.
I don't think these use cases can be served by the same system. The whole systemd schism was a symptom of this core problem - systemd is great for the laptop/mobile/embedded case because it adds a whole lot of control over various system states. The server guys hate it because it adds a lot of complexity and doesn't give them much functionality that makes their lives easier. But the problem isn't systemd, its that these two use cases are very hard to solve with one system.