However complicated this solve looks, getting it working was so much worse. The main challenges were separating the quintessence and output assembly.

The big breakthrough for quintessence was done by taking advantage of arm 2's two clockwise rotations. My failed attempts used the second rotation to move the quintessence, but by switching to the first spin, the glyph of dispersion was blocked by the fire held in arm 2. This block allowed arm 6 to grab the still-whole quintessence before arm 2 finished its second spin (cycle 55 for example). Staggering arm 2 like this single-handedly makes the solution work.

The assembly was easier than expected. Fortunately, the second quintessence arrives just as the first triplex bond is formed. This timing allows me to conditionally output with 4 instructions, and store only one quintessence.

Overall, I can't tell if this is a local minima or if I found the sauce, but any further instruction shaves would need a different approach for dispersion.