Video decoders often use fixed-point optimized iDCT approximations to increase speed at a minimal quality loss.
I'm sure some video decoder pro can explain it properly, but video decoders just don't work like that. They output the same pixel format that was fed in. If it would be beneficial to increase bitdepth on output, i'm sure someone would've come up with the idea before..
__________________
LAV Filters - open source ffmpeg based media splitter and decoders
|