Yes! Haha! Yes! I fixed Tangerine's occlusion culling tonight. While this isn't the silver bullet I was hoping it would be, it does substantially improve the draw time of my worst models. The most extreme case brings one of my interpreted models down from ~83 ms to ~30ms.
an idea: if a given voxel hasn't moved or otherwise changed shape since the prior frame, the culling test could be made stricter to eliminate false negatives for empty voxels in front of the actual rendered surface 🤔