The answer to this question actually comes in two parts.

The issue behind my MapOptGenToRegions seems to have been related to the actual problem set up. Because of how the declaration of a thin film interference layer relies on a ReferenceLayer to work out Reflection, Absorption, and Transmission, for the purposes of optics, the stack must work like this:

Gas

ZnO with gas' complex refractive index: call this material dummy_ZnO

CdS

Then ZnO is declared as a VirtualRegion. In the RayTraceBC( TMM (...) ) declarations, the ReferenceMaterial is 'dummy_ZnO', and the thin layer is the true ZnO material. This is an obvious hack, but the seems no other way to have the CdS 'in contact' with something which is optically gas, but electrically in contact with ZnO.

All of this works only if one assigns a non-zero (even if extremely small) imaginary refractive index 'k' to the dummy ZnO. I suspect this might be related to a floating point exception which is generated under the condition of total internal reflection, but I can't confirm that.

All of this is still a ridiculous workaround, and declaring ZnO as a virtual region should make the OpticalSolver understand that CdS is optically 'in contact' with Gas, with a ZnO TMM intermediate. But it doesn't work, so the workaround appears necessary.