That -Z suffix always intrigued me. And obviously, it also was the last reimplementation of the PDP-11 without using a microprocessor.
I assume they started seeing that that the microprocessor was the path to the future, and the 11/44 was just at a time when it wasn't yet there, but another two years, and it would be.
They needed to get something out there that fit in a corporate cabinet, had decent performance, and was reasonably inexpensive. They were still selling 11/70 systems although they didn't particularly want to by that point (having done the DECdatasystem 570 because PDP-11 was a dirty word in some parts of the company and user base). The 11/34A didn't have the memory expandability and the 11/24 was stretched to the limit, performance-wise.
Another was that they were probably a bit tied up with trying to get the VAX on a chip, so in order to make the PDP-11 happen they were more or less forced to find some external company that they could collaborate with. The VAX was a higher value and importance, so it would take priority when it came to using internal resources.
I expect there was some internal horse-trading of the form "We'll let you do the design work for the J11 if somebody else does the fab work, if you let us use your FPA design".
I think it was partly because DG didn't really have enough resources to go with developing a microprocessor. I bet they would have loved to do it.
Although I was well "out of the loop" by the time the S/140 came along, I was surprised because the earlier Eclipsen were all '181-based and shared a mostly-compatible WCS implementation. I expect that the S/140 either didn't have a WCS option or it wasn't even "culturally compatible" with earlier models. But as I mention below, WCS was always a niche product.
DG was more interested in chasing the desktop/deskside segment at one end (MPT/100 and MP/100 originally), and the "big box" at the other end (like the C/330 and M/600). They seemed perfectly happy to continue to sell the existing midrange with only secondary improvements like semiconductor memory and later the ALM communication chassis instead of serial ports in the processor chassis.
That did give them time to develop AOS which became the foundation for AOS/VS on the MV/8000 family machines. The MV/8000 series was a stopgap that eventually ended up carrying the company over until they could switch to Unix-derived systems.
I think the interesting discussions would be more about the 9000. The 86x0, while delayed, and initially underperforming was still successful. The big reason no 8670 came about was really that DEC had started to want to lock down the buses, and the 86x0 just didn't fit in that scheme. The BI bus obviously became like a mantra after the 8600. So the 8650 was allowed to be done, since the effort was rather small (as you observe), but by then DEC was transforming into a company that really wasn't that interested in being an engineering company anymore, but which wanted to be IBM. And in that picture, something like Unibus, Massbus and SBI just was not what they wanted.
Yes - the era of "Let's hire 10,000 IBM sales people". Who unfortunately were a lot better at convincing customers to buy IBM than DEC because they were more familiar with IBM products, despite handing out DEC business cards.
And DEC completely ignored the lesson of the benefit of a thriving plug-compatible ecosystem for the IBM products. Without which DEC would hever have had the RP07 and TU7x peripherals.
The 86x0 came with a way-overspecified 60A Russelstoll plug so customers could "just" unplug and wheel out their 370-architecture systems and roll in a VAX 86x0. Never mind that everything else - hardware / software / operations were completely incompatible and different. The closed thing to compatibility was that IBM and DEC both offered a RPG II compiler.
The 9000, when it came out, was only a few months ahead of the NVAX. And the first iteration of the 9000 was not as fast as the revised one, if I remember right. And considering the size and price of it, it should have been obvious that they were only ever going to sell few of them. And microprocessors were increasing in performance so fast it was only a question of time before they would catch up and surpass.
I don't know if it was the sunk cost fallacy, ego, or total ignorance/disbelief of where microprocessors were headed that perpetuated the 9000 long after it should have been killed off.
To make another comparison between the PDP-11 and the VAX. The 86x0 was in some sense similar to the 11/60 on the PDP-11 side. It had issues, were late, and never realized its full potential. And came about the same time through the curve of the product. Admittedly, the 11/60 was in the end much less successful than the 86x0. Maybe just because it became much more crippled compared to the high end PDP-11s than the 86x0, which was closed to bleeding edge in performance.
The VAX architecture had the benefit of being fully specified. Later models had the ability to subset it (like the MicroVAX II) but actual architectural extensions were rare. There was VAX Vector supplemental instructions, for example. I don't remember if SMP introduced any new instructions at all, likewise for VAXft. Whereas the PDP-11 architecture was the Wild West by comparison. That meant that machine-specific extensions could be tried and if unsuccessful, omitted from future models. Extended addressing, I&D space, supervisor mode, cache, WCS, CIS, etc. Some of these caught on and showed up in future models, some were architectural dead ends. Unfortunately the 11/60 was late enough that other machines (like the 11/70 and to a lesser extent the 11/34A) beat it to market and stole a lot of the thunder the 11/60 group apparently expected. Aside from WCS, it seems to have been an evolutionary dead end. And WCS was only of interest to a very small percentage of customers, particularly when it became obvious that it was a model-specific feature unlikely to be available on subsequent models.