wth is this example even supposed to portray? It's far off any reality.
It's completely nonsense to think chroma would be blurrier than with bilinear.
software decoding + bilinear:
DXVA2 native decoding + bilinear:
DXVA2 native decoding + DXVA2 scaling:
Oh, and bilinear downscaling also gives you one aliased mess for every target which is not exactly 1/4 or 1/16 etc. of source resolution btw...