.NET was already a going concern before Longhorn even started. What sank Longhorn was the fact that writing an OS from scratch is hard and maintaining compatibility with existing OSes in the process is even harder, especially when you're adopting a completely new architecture. Longhorn would have been a microkernel running 100% on the .NET runtime, mainline Windows is a monolithic kernel written in C++. I don't know how it would have ever worked, whether .NET was "perfect" or not.
No, Longhorn was neither a microkernel nor was the kernel rewritten in .NET.
Source: I was there.
I think he confuses longhorn with Singularity research project.
See Android, or Meadows for alternative reality.
Android still runs on a monolithic kernel written in a memory-unsafe language. I'm finding it suprisingly difficult to find information on Meadow, other than it runs .NET DLLs as user-space applications, but nothing about the structure of the kernel.
Longhorn was going to be more than that. Microsoft did have Singularity/Midori projects, started around the middle of Longhorn/Vista, and continued much longer after Vista released to build out the managed microkernel concept. It's been about a decade since they've put any work into it, though.
Microsoft wasn't even able to deliver that, which was my whole point.
Joe Duffy mentions on a talk, that even with Midori running production workloads, Windows team could not be changed their mind.
Meadow uses a C++ based microkernel, the whole userspace is based on .NET, by the way.