“Due to DX11 hardware limitations, the cubemaps used to capture the scene are all 128 on each side”
Does that mean its 128x128 texture resolution for each plane of a cubemap (6 in total) ? there is really no way to improve it ?
By the way, you seem to be saying that spheres reflection captures are more often the way to go, but I have the feel that if a box reflection capture is used correctly, it can give better results and fake matching up that spheres rc (for having tested it myself)