AppleScript would be one approach here. Since I don't have time to actually implement this right now I'll give an outline of what I'm imagining in case you or somebody else on here wants to take a crack at it. Or maybe there's a simpler solution:
Write a script that looks for untagged detected faces and creates an album (but not a smart album, unfortunately) with all of those pictures. It could delete the old version of the album before beginning so that it behaves mostly like a smart album, except you'll need to run the script to update it. It might be possible to set it up to automatically run on some trigger.
To make things complicated, iPhoto '09 doesn't expose the Faces feature through the AppleScript library. So you'll have to open the sqlite database within the library bundle and read that manually. Somebody has already done some of the work involved there and created a script for interacting with Faces.
But those scripts are parsing the faces that are already attached to a photo. We're interested in the potential matches. The
faces.db sqlite database has a
similar_faces table which appears to list all the unconfirmed faces that it has detected. You would want to select the
image_key for all the rows in that table with
ignored = 0.
There are sometimes multiple faces listed for a single face in a photo. Those possible matches show on the page for the face ("John Doe may also be in the photos below"). If you're only interested in faces that iPhoto is confidant enough in to suggest a name straight from the picture, then only select the rows from
score greater than some threshold,
0.80 or something. You might also need to look up the
face_name to make sure that the strong match is to somebody that it already knows about, I only have two untagged faces to examine right now so I'm not sure.