I can see the misunderstanding. I didn't claim they were kernel - just that perceived problems with Windows were seldom actually kernel issues. At least directly.
Kernel-level drivers, with bad or non-existant threading models, mechanically executed ports from earlier Windows versions and crappy error handling. That has polluted the NTExecutive kernel - an otherwise admirable piece of software.
The modern Windows kernel is much improved from the original. Design ideas around NUMA that hearken to Cray and SGI are present, and support multi-core in remarkable ways, that differ from the old SMP problems. The support of virtualization is also an ESX rival.
These ideas were also cleverly pilfered. The most unethical sorts of things happen at MS - and I happen to know of at least two competitive marketing-funded operations, which could be called "dirty tricks" or "black bag" teams.
There have been brilliant technical leaders in this space beside Cutler - who has been at pasture for years. Bill Laing is salt-of-the-earth. I am fond to recollect the few occasions I had to meet him. Mark Russinovich is also a certifiable genius, and responsible for influencing the additional capabilities in NTExec in impressive fashion.
We have 25 years, passing from those old days. So? NT Exec was more-or-less VMS. Shall we say a strongly typed derivative? Those who say they don't see the resemblance, are like those who don't know why OS X is like BSD. ;-)
This brings up another comparison: Modern Windows is an updated, VMS legacy branch in much the way OS X is an updated, NeXTStep branch...