This is probably the best "coaxial" output shape I've had. Since the simulation is affected by the mesh, I decided to simulate the full geometry instead of using a symmetry constraint, and it only took about a minute to simulate but my residuals were much better (sub 10^-8). 

A diamond square isn't ideal but at least it's on brand and colours should be consistent on all 4 cardinal directions (e.g. the XY faces on a calibration cube) which are probably the most common.

I had to shuffle some things around but the result looks decent:

I've broken the sub 11 cm^3 barrier and now at 10.5 cm^3.

[Aug 23] 

I've changed the output to transition from a square to a circle to eek out a bit more circularity:

I believe the outermost channel being circular is the most important, as it will be the one to affect interlayer adhesion the most.