It should be possible to have the PPU emulation capture all of the final register state per pixel (or scanline if accuracy isn't paramount) and have the GPU render each pixel using only that state, doing the layer blending, color math, and mode 7 calculations as necessary. Based on MVG's video breaking down the draw commands performed it doesn't look like that's how Super ZSNES have implemented their PPU - it seems to render tile by tile for BGs (and OBJ?) and line by line for mode 7. That'll be a bit inaccurate but it's likely necessary to implement some of their visual enhancement tricks.