Wouldn't it be easiest to keep the original retarget algo (with 2016 block window etc), but just do the retargetting more often so that retarget windows overlap? You could retarget every block if you wanted (or every 10 or whatever).

With system like this, for example Namecoin difficulty would have dropped to "profitable" levels a long long time ago, and it would have suffered only a temporary slowdown.

Is there a downside to a system like this that I don't see? Some problem it doesn't solve or some exploit it introduces? It seems so simple :-)