[1] Around this time, he reported to an executive named John M. Thompson, hence the use of middle initials.
It's another thing to pretend he's a robot and has no feelings on the matter, which is what you're implying. I think the original question is a valid one. Clearly he would have thoughts on the matter and they may be interesting to hear.
I know nothing about his career or his decisions WRT the OS/2 product, but, after having a closer look at the other two, a pattern of executive placement as an offensive weapon starts to form.
edit: minor clarifications
It's far-fetched anyway. Belluzzo is much less far-fetched and Elop is pretty obvious.
Good Lord, that is some serious speculation :)
Unfortunately, per Bill Gates, its 286 was "brain dead chip" and absolutely hopeless and slow in Protected Mode, which is where it needed to run for a modern, large address space computer. As I understand it, the OS/2 inability to abandon the 286/PC-AT for the 386 is what really drove the split.
(It wasn't even IBM that came out with the first major 386 system, but Compaq.)
What I was told by various people who tried this is that Intel had no idea how people actually coded for the x86/weren't willing to make segment sizes larger than the original 64 KiB, and the penalty for loading a new segment register was "tens of cycles" per Wikipedia, I've heard on the order of 40 earlier.
So a lot of existing code bases, including a couple I worked on for PCs (Gosling Emacs and a write once filesystem) ran, or would run, fantastically slower in Protected Mode.
Yeah, sure, you could with no small amount of pain write more efficient code for OS/2 1.x, but why bother after Windows 3.x became a big hit?
Oh, yeah, now you're reminding me that the cost to develop for it was $6K for the SDK in 2013 dollars, plus of course a PC-AT with a lot of memory (at least 1.5 MiB, which wasn't cheap back then, nor was the AT).