I am doing similar things but decided to start from debootstrap woody instead of KNOPPIX as it would be much more difficult(for me) to remove things from KNOPPIX than add things to woody. Beside, it seems that the overall binary size from sid/sarge are bigger than woody(could be the compiler used).

I have managed to sqeeze a bareX with quite some server modules within 300M(uncompressed) and a base woody system within 150M or a KDE 3.1.3 based system within 400M.