Personally, I wouldn't be stoked to run Linux on them either to be honest. But both are being done. Practicality rules I suppose.
With Windows, you need to beg and pay Microsoft for customizations and hope these changes will not cause other issues.
Plus, most space projects are on a tight and limited budgets where management would rather spend on hardware than software.
But what would you run? QNX? BSD?
A better way to put my argument is: could an average mom build Linux on specialized hardware in space? If the answer is "yes", then you may have a point.
I don't think the answer is yes.
Just because you can read it, doesn't mean you can or will be able to actually fix it; not because of technicality, but because of personal knowledge.
In this case, they are both black boxes.
I once spent three days trying to figure out an issue, stepping line by line through hadoop (after figuring out the issue was in hadoop and not my own code). Yay, I proved the issue was actually in Java itself. Guess what happened next? We avoided the bug. Why?
- We couldn't update Java.
- We couldn't change hadoop because we were using a packaged solution. So, we just filed a bug with them.
Had the source not been available, we would have just skipped all of that, and it would have been our vendor's problem 3 days earlier.
This happens less nowadays, because stuff like Android and ChromeOS, so why pay Microsoft when free beer OSes exist.
If you look at an LTS branch, you'll see there are hundreds of point releases. Usually a point release is created once every 5-10 days. I interpret that to mean bugs were not found until many weeks after the LTS branch was cut. Obviously, not all of them affect you, but many patches apply important subsystems which affect you.
That said, this is on the scale that I'm surprised off-the-shelf things are even being considered. I'd have thought they'd just roll something bespoke and that's the end of that.