Well, in my quick test, I used the default setting of a keyframe interval numerically equal to the frame rate (to produce 1 keyframe/sec), in this case, 12. The result was markedly inferior to the MJPEG result at nearly the same bitrate (identical source material). So, my tentative conclusion is that the result has more to do with the particular way in which Cinepak performs its vector quantization (and motion estimation). Since Cinepak was designed for easy decode on CPU-challenged machines, the algorithmic tradeoffs must have been fairly dramatic.

Perhaps by fooling around with the quality slider and target bitrates, etc., (or by using a better Cinepak encoder than the one bundled with MoviePlayer 2.5), I could achieve a very different result.