Well, I spent whatever free moments I had today to launch a series of transcoding experiments. And even at max quality, Cinepak loses to MJPEG at equivalent bitrates. In all cases, the source clips were NTSC VCD (352x240), 24fps. The output clips were 10fps, 12fps and 15fps, all at the same resolution as the source. Keyframes were kept at a constant one per second. The target output bitrate was 700kb/s (chosen so that one might fit a ~2hour movie on a single CD).
One source clip was from Galaxy Quest, as mentioned previously. Another was a comedy bit from Brian Regan. In both cases, Cinepak produced output that was markedly inferior to MJPEG, in the not expert opinion of every single one of my coworkers.
Aside from very noticeable motion artifacts, Cinepak also seems to truncate the color space so much that human faces can have a comic-book appearance at times, depending on what else is going on in the scene. From reading what others have found in the course of more detailed experimentation than I've run so far, this is a very typical result.
So, in the absence of additional guidance on what, if anything, I'm doing wrong, I conclude more firmly than before that Cinepak simply produces output inferior to MJPEG, at least for these framerates/bitrates. That's not the same as declaring it useless, however. As mentioned already, the light decode requirement makes Cinepak very attractive for video on machines that otherwise can barely handle browsing, for instance.
Btw, the encode speed for Cinepak was 2x slower than for the MJPEG cases, so Cinepak appears to be very asymmetrical in its encode-decode requirements. On a 300MHz G3, a 4.3 minute Cinepak clip took over 2 hours to encode! I shudder to imagine how long it will take that LC to finish its job. Wow! It might take a day or more!