[leguminous]

I have had the same thought. Why not do away with chroma subsampling and just compress the chroma planes more heavily than the luma plane? Does heavy compression perform worse than just throwing away 3/4 of the data?