David,

My experience is that the encoders will handle motion within the frame far better than camera motion, but that being said, the quick movement of martial arts will 'stress the codec'.

Codecs like H.264 and On2 have motion estimation code that tries to recognize either subject motion or camera motion and save bandwidth by finding similar blocks of various shapes and storing them as the same image. This is all part of the black magic of writing a codec.

For example, with the x264 codec, you have a lot more control over the encoding parameters than most other H.264 codecs, like tweaking the way motion is detected, or subpixel motion estimation.

To really understand these settings, it's necessary to do short experiments of various types of content and visually compare the differences between the encoded movies. Because of the exhaustive nature of this process, I wrote a software environment called SampleLab that makes this kind of experimentation simple and efficient.

But, to oversimplify things a bit for practicality, you could say the more pixels change from frame to frame, the higher bandwidth you'll need to preserve the same quality.

For a detailed description of the tradeoffs between codec, bitrate, frame size, and content motion when encoding web video, I would recommend studying pages 16-22 of the DV Kitchen manual. You can get this PDF manual for free by downloading the free DV Kitchen trial. (OS X only at this time, sorry)