I seem to remember that bogofilter only looks at the first x number of characters (or was it lines?) in an email. Is this true of both registration and classification? I'm thinking of a simple optimization of truncating mail in the training corpus to save space, cpu cycles and bandwidth.