yes, ffdshow as far as i've been able to see is the fastest directshow decoder.
mplayer is actually twice as fast, though they are both using ffmpeg to decode. i put the extra speed down to directshow bloat.
to clarify - i've done some encodes of field-blended stuff bobbed up to 50fps, 512x384, plus vobsub subtitles (it's anime, and i hate english dubs).
xvid+directvobsub choke on these videos - i'll get a gillion freezed frames, and average decoding rate is about 30fps when it's not frozen.
ffdshow+directvobsub performs a little better - about 40fps.
ffdshow+internal vobsub decoding gives me ~49.998fps, with the occasional frozen GOP. i consider this enough performance to be used.
mplayer+directx output (mplayer accepts vobsub) gives me flawless decoding at ~48% CPU usage.
mplayer+null output (as a benchmark) gives me a staggering 30% CPU usage.
i'm running a p3 733, geforce DDR 32 (old, old), and 384 RAM on win2k.
i haven't done a speed check under linux, as all i have is a live CD distro, and haven't got the hang of it. i'm willing to bet it's faster under linux, though

it's amazing what you can squeeze out of a ~5 year old machine.