One of the goals with these amps was record setting low distortion for the time. The front ends were all discrete using choice Japanese parts, and relatively advanced stuff like cascoded and/or a dual differential complementary VAS that served to drive up the component count. Boosted rails were used, Baker clamps were employed. Suffice to say they threw a lot of good ideas at these things that were far outside of the norm. It wasn't long before they learned their (probably expensive) lesson and started sourcing trash amps from overseas.
Given the quality of the rest of the design, my assumption that driving the output stage with no less than EIGHT fast(ish) drivers per channel instead of TWO slow drivers probably had a tangible benefit. This stuff about production costs is interesting, but it makes no sense in context. It takes more space, more screws, and more time to bolt on four MJE drivers instead of pushing the whole stack of output with another big, slow MJL driver, which is what almost everyone in the sound reinforcement industry did and still does when they start pushing a hungry stack of MJL workhorses.
Unfortunately, there just aren't many amplifiers using more than 4 or 5 paralleled outputs at high voltages. As good as, say, Doug Self's observations about reducing distortion by just chaining up more outputs are, you do run into a practical limit of how you drive those outputs if you want to use good old 2119x parts.
My original question/observation, which perhaps could have been more clearly stated, was simply whether using an arrangement like this is superior to an arrangement which uses another 2119x to in turn drive multiple paralleled 2119x outputs. I think we went a bit far afield.