Sorry, but I'm not totally following your thinking. If you apply the Lambert cylindrical equal-area projection in that rectangular texture you will only get some distortion near the equator line.
And that's the whole point: 'some' distortion. I'm trying to minimize the polar distortion as seen in unprocessed textures. And therefore the pixels towards the poles have to shrink. I know that it will still look distorted, but way less.
Pretty interesting that we get an almost undistorted projection when we first do the sinusodial prjection and afterwards the "rhombic" one. But as we found out, there is too much loss of texture quality (especially with no supersampling at all).
Something that came into my mind while reading your last paragraph is that the original output from Noise2 is a square image. And thinking over this whole subject i realized that we've been working with rectangles the whole time. "Reconstructing" an imaginary square is impossible, too. Now what happens if we shrink down the square to half its height, but not dividing it by 2. Instead a sine-function comes into play. What do you think of this?
something like this:
<img src="http://dl.dropbox.com/u/6200498/moon.jpg" border="0" />
polar distortion is there, but less than with an unprocessed texture.
furthermore... thanks again for your time, and sorry for me being so neurotic on this topic <img src="smileys/smiley23.gif" border="0" align="middle" />