How do we select the luminance threshold then in reality? Do we actually have to look the rendered image or there is some metric to tell us?
At this point, the image is starting to have some noticeable artifacts, even without comparing it to the original. Was it really worth saving only 1/3 of the time to have an image that's this much worse?
Are the shadows always the first part of the image that become more emphasized? As the artifacts increase, it generally looks like the lighter areas tend to remain more accurate to the original.