Let me see if I can illuminate from a little I was aware.  I know some of the histories from my time at Locus when DG was a customer of ours.   Although, I'm going to show my old f*rt, cultural illiteracy (not a TV person) by having no idea who 'John Snow' is, so I can not comment on that reference ;-)

I think you are right that some of the histories tend to tided to people's experience at their college or universities and I'm not sure how well DG sold in those markets.  Plus they bet on Moto, who fumbled with the 68000 replacement (88k), compared to IBM's PPC and later Intel's remarkable recovery with the 386.  Since they joined the losing side of the chip war, I suspect that also hurt them, even though the system was actually well done.

At one time I had access to the sources of DG/UX and yes you are correct it was very clean and easy to understand (and fairly well documented).   At the time, we (LCC) had the sources to DG/UX, Ultrix, Tru64, OSF/1, SunOS, Solaris, HP-UX, AIX, Prime, as well as all of the AT&T releases.  (The management at Locus used to say we were the Swiss of the UNIX industry  - we sold arms to all the sides of the war).   

As for  DG/UX, their kernel seemed like it was a rewrite (I never knew how much of it was the SW folks in NC (who had been building their failed "Fountainhead" system that the Soul-Of-The-New-Machine book talks about, and how much was the MA folks that did Clarion later on).  It was definitely the sanest of all of the UNIX kernels we had access and had the least amount of cruft in it of the commercial systems.  The locking scheme was the one of the cleanest, I ever saw (Stellar's Stellix was the only one that was as good, IMO).  The memory system was really impressive.  I remember when we were doing the TNC distributed FS work that what would become the TruCluster FS for DEC at the same time.   The DG/UX version was the simplest (and the HP/UX version the most twisted confused). 

I'm now mixing up the differences in my mind, but I think I remember DG/UX had some sort of file system stacking scheme at the inode level.  I'm not sure if it ever shipped, but we worked on a Union file system for them; that I remember was very cool (Plan9ish in splicing file system namespaces together).   I seem to remember it was all made possible because of the lower level memory scheme.   But I might be mixing that up with one of the systems we were hacking (it was definately not OSF/1 or its child Tru64). 

As you said, the user space API was basically System V, but DG did support a lot of BSDism also; so bringing networking code from 4.2/4.3 was not terrible; although since it was not fish or fowl, you had to be careful.   The big issue when it came out, is that SunOS (and later Solaris) had become the defacto system at most universities, where much 'free software' was being produced.  Since it was more System V user API at the heart (which was good for DG's targetted ISVs), it did suffer from the porting issues.  Since it was an 88K and SunOS had been 68K, the Endian issues of the free SW was less of a problem, but not only was not a VAX, but it was not BSD. 

Clem