Well for xvid files atleast, if the file uses nonstandard packed b frames, where the b frames are muxed together with p frames on the container level, tagging them on to the end of the p frame, the codec will see this big p frame, and complete decoding of it, and discard the extra unrelated data (in its opinion) and wait for the next frame to be fed from the demuxer...

The result being all b frames getting dropped, which can be as much as 2/3rds of all frames..

I guess the solution would be to either make the avi demuxer aware of it, and teach it to look inside and parse mpeg4asp frames, reorder them and feed them properly to the codec, or to make the codec aware of them...