This solution is probably as close to a magnum opus as I've created thus far. As a RI puzzle, I felt like this was most approachable for a less-experienced solver like me who isn't as familiar with the well-explored tech of this game, as I could take as much time as I needed with crappy designs as long as I got the optimal rate and kept the period low. I committed to the lowest practical period of 4.

I had initially thought the optimal rate was 7/72 but realized I could squeeze out a little more halfway through the implementation. Going back to the drawing board, I re-derived an optimal rate of 15/136, but the path to it was far more treacherous: With 136 reagents, create 119 copper and 136 quicksilver, upgrade 45 of them to silver, double upgrade 16 of them to gold, and do a combine-then-upgrade to the final 58 of them to gold. Multiples of disgusting prime numbers. 

And thus this giant machine was born, creating those prime number chains the only way I knew how: manually count out every. single. atom. It may not be optimized at all, but it's something I can be proud of. Thanks to those who helped me figure out how to install mods on my Mac, none of this would be practical without you guys. And thanks once again to zorflax for piquing my interest in this little game I played once upon a time to a whole new level.

Some points of interest:
1. The huge swing of what would become combine-then-upgraded gold narrowly misses an arm by a single cell grid.
2. The machinery that handles the quicksilver used for the final product magically also gets reused to transport the silver.
3. The parity of how quicksilver atoms get bonded changes in a three-cycle since 136 is not a multiple of 3, so the downstream machinery has to conditionally handle all three possible states.