The cinmaster engine(cinmst32.dll)
is the fastest so what I heard.
You can see lot's of mmx blocks
when you disassemble the dll.
I guess the other engines do the same thing.
The graphics card plays a very important
role in the video decoding too. The ATI
players are paired with the hardware
accelerations(motion compensation,
IDCT and color space converstion) on ATI cards.
It's really hard to tell without a specific
system configuration.