First Design because i couldn't get anything better than it, made by slapping down arms 7 & 8 for the N=3 stick input, and then the multibonders. Working backwards from the final input, 3 every 6 pipelines for 1/3 of the output seems like the obvious way to go here.

Needing a trackloop at the top feels kinda bad, but latency on the N=2 stick input is still tight and geometry is hard. Water routing to the top left is also a little sad, taking 50g. 

The rest of the solve feels pretty good however, working on a clean every-6 loop. Guessing min G for 39C will be about 100g cheaper than this at about 350g-400g if you can fix my problems outlined above, reuse some bonders or not have 2 convenient expensive multibonders in the middle, and save a couple arms.