I think you might be getting hung up on the film equivalent. For example, when doing a multi-exposure shot with a film, once you blow a highlight it doesn't matter what else you do, that section of film is blown out. So if you were to take two exposures (lets say 0.7 secs for the shadows and 1/45 for the highlights and keeping the aperture constant), it wouldn't matter, your highlights would get destroyed when exposing for the shadows. If there is no motion, this is basically an addition, the equivalent of exposing the film (or a single picture) for a length of 0.7 + 1/45 = 0.722 seconds. So for film, no matter how you pick the exposures, you will not get any increase in dynamic range captured. And Film multi-exposure is an addition of each frame.
But there is a difference between a film camera and the K20D. The K20D has two multi-exposure modes, one that emulates the film version, and a new one that is only possible with digital (AFAIK). The mode is selected by the "auto-EV Adjust" checkbox. If this is not checked, then it is like a film camera, and you get no increase in DR. However, if you do select it, the camera doesn't add the exposures together, it AVERAGES them. So even though one image has the highlights blown out, the second has the highlights well exposed, and the net effect when you average the two is that you maintain highlight detail, but the DR of the highlights are half of what they are in the well-exposed image. Same goes for the shadows. The key is the average, and it's not an addition. So the resultant image will be like this: the portions of the image that are well exposed in both frames maintain the same DR as either frame, and the portions that get over-exposed/underexposed in only one frame will have half the DR compared to the well exposed frame for that highlight or shadow.
You've maintained your detail, you've just compressed the true DR of your scene into the 8-stop range that a JPG allows.
I also took some test shots to demonstrate this. I lit a portion of the frame (my bathroom) so that it metered at 1/45, and lit another portion so it metered 0.7". This range is larger than the DR of the camera (at least for JPGs). So for a single exposure at 1/8, both highs and lows get clipped:
Now, for the multi-exposure I used two exposures, 0.7" and 1/45:
The camera combined them with this result:
Compare this the the 1/8 image from above. I think you could say that now the dark sections appear up reasonably well exposed compared to the 1/8, and the bright areas (the shower curtain especially) also has more detail that the 1/8 single exposure.
And to round things out, here are the histograms for the images:
I apologize, the middle plot on the left is mislabeled, it should say 1/8 (instead of 1/20). I didn't feel like re-generating that plot. You can see that there isn't a single exposure frame that can cover the DR of the scene, and that the 1/8 exposure clips both highs and lows. But the multi-exposure image has a nice looking histogram, with very little clipping. The color of the wall behind the shower curtain does look like it's changed, perhaps I should have manually set the white balance, maybe thats what makes it look funny.
Anyways, hopefully that helps.