Regarding AIX I tend to disagree.
I was contracting with IBM and helping them out in HACMP and PSSP support back in those days of AIX 4.3 - 5.1.
Even though we managed to blow a computing center fuse once by powering on 80 fully loaded power 3 nodes with 16 cores each simultaneously, booting, recovering, ..., etc was not a big issue. Neither was reinstalling the OS. One of my colleagues and myself wrote a script to do that on those 80 nodes including setting up partitions, rvsds and GPFS plus system tuning in less than 4 hours. That's reinstalling on 80 nodes including cluster setup. Try to do that with Windows ...
Also the kernel was easy to read, we had this nice serial entry point via console and NMI, and could clear a hang via direct manipulation of the cpu registers (including bumping hung threads off the core it ran on). The only situation where we were completely out of luck was when part of the kernel was paged out and interrupts were disabled.
I once remembered a session in one of the internal training classes (where I usually would not have been allowed to participate, being an external), when the presenter showed us the various interrupt handling calls and I asked him where lock was freed as I was only seeing the interrupt handling code. He looked at me for a few seconds, then started rummaging through his code until he finally found the call to the assembler locking instructions. The guy sitting next to me (one of those AIX champions) said: how come you know this stuff ? you don't have access to the source ? and I told him: why would I need that, I have a disassembler ... *dead silence*
Those were good times, in which real men were still real men and real women real women. *sigh*.
TE involved: kernel debugger ...
BTW I still have that 4.4 BSD source license ...