Somehow I keep finding small optimization on this thing. I've submitted it several times, each time thinking surely this is the final form, and then a couple days, or sometimes a couple hours later, I'm proven wrong. Perhaps it will be the same with this version as well.
The main change from my last version was swapping the bonding site in the brick from the middle lane to the bottom, thus allowing the bonder in the final construction to pick up pieces as they detach from the brick. This severely limited my ability to move the brick without losing rate, so I wasn't able to get as much of a latency reduction as may be expected, and in fact the one I did get only came by moving the brick early at the end, and thus breaking the loop. It did however save me 50 gold in the form of 2 arms and 2 track.
It also gave me a very interesting time with re-establishing a loop. With one more track piece for arm 3, it's actually trivial, but when properly optimizing all 3 metrics it was a fair challenge. I ended up not finding any clever way to move the remainder of the brick to re-convene with the new parts after the loop without breaking in the construction site, so instead I took the next most rational option: taking apart the entire brick at the end of the loop. Perhaps not the most satisfying way to solve the problem, but it loops now.