Yeah, the RGB image looks really wrong.

The CMYK image, interestingly enough, gets close to NASA's "Official True Color of Mars" recipe. Given my skepticism of NASA's recipe, I got a chuckle out of your observation "As you can see, it's totally off and looks almost one-color." Well, I think NASA would actually approve.

My initial hypothesis about NASA's recipe was that they colorize monochrome images to a hue value of 19, but I've since begun to refine the hypothesis into a recipe involving heavy reduction in blue, moderate reduction in green, and heavy reduction in contrast in both channels. I was playing with that just last night.

Your description of how you did it is interesting because you don't mention manually adjusting the blue and green channels. Inverting and swapping channels seems appropriate, because the cyan, magenta, and yellow in CMYK are the inverse of red, green, and blue. I think the key may be that you used L2 for black, which I think would result in the light areas of the image reducing luminance.

I started playing with CMYK for the first time this morning. For reference, here's the L4_L5_L6 RGB:

Here's what I got with C=L4' (' = inverse), M=L5', Y=L6', K=L2':

I'm intrigued by the higher quality of this version. I think that it's because the L2 image in black more precisely "modulates" luminance than simple RGB combination. I'll definitely pursue this further.

Since I've almost never done CMYK, when I first started I screwed it up, and made a serendipitous discovery: When I do it exactly backwards (C=L6, M=L5, Y=L4, K=L2), I found I that I can recover data lost in the shadows:

Cool, eh?

Thanks for getting me interested in CMYK (SAIDias too). This could turn out to make a significant difference in how I process the images.