complex issue.
down in the engine you could use a stencil operation to limit each scene render in the raster stage. you’d have to prefill the rendertarget or the main backbuffer with the “splitter shape” and a simple stencil value of 0 or 1.
when rendering each scene camera view you raster only what is stencil_ref 0 or stencil_ref 1 per respective view. you still render 2 views but should have a single pixel output for both segments, no overdraw.