Bruce, I don't think you are going to find hard "evidence" for either
conclusion. But Bruce's conclusion is consistent with my own
experiences, and that of many other Administrators that I discuss
issues like this with.
Since its inception, supporting NT 3.0 beta and onward, I have been
dealing with BSOD's. In total, there have been comparatively very few
times were it was a direct fault of MS code. It has very commonly
been in relation to 3rd party drivers that needed reworking or
updating by the 3rd-party manufacturer.
This is not PR spin (of which I don't think you could find any
published PR spin for either side of this argument either). This is
real world experience with the NT+ products across i386 and Alpha
hardware platforms using peripheral devices from many different major
manufactures. There are admins on both sides of the anti-MS fence
that I communicate with that would agree with this conclusion.