The complete opposite is true. @skw gave a good list of counter-examples, but it’s far from exhaustive. Remember when Java was the next big thing that all Mac OS applications were going to run on?
Oh, okay, we’re arguing about what “big” and “small” is. If one thing is six times bigger than the other thing, that other thing is small to me. But that’s not important, the point is, is it big enough?
Not at all. Like I said, a huge amount of software doesn’t ship on Mac, but also a lot of software is Mac-only. Supporting multiple platforms has to pay off. The actual market share of Mac users for your niche may be 95% or 0%, depending on the kind of software you offer.
I wholeheartedly disagree. In an abstraction layer, when you add some feature and it requires something your layer doesn’t cover, you need to touch N+1 places. Sometimes one feature isn’t available on all platforms, leaving you with the lowest common denominator, or a leaky abstraction.
You also have the mental overhead of how your abstraction layer translates into actual platform code for N platforms. Refactorings become more expensive, the architecture will be more rigid and less likely to improve.
Last but not least, this is the kind of work that absolutely nobody likes to do, which is bad news for a project like Blender, which is supported by volunteer and below-market-rate work.
No developer has stated anything to that effect, as far as I’m aware. You should read the whole thread. Ton has signaled interest in MoltenVK, but he isn’t keen on acting right now.