The problem with weighted averaging such as how Mike combined multiple prints into one, is that it leaves you with merely the ‘average’ quality of the initial prints. I think this is what is resulting in the plastic look of the people in the speeder shot. Say that we have at our disposal a sharp print and a blurry print. If you were trying to resolve a blurry boundary between two objects in the frame into a sharp line, then (assuming we are only gathering data from the same frame of both prints) averaging the values of a blurry print with the values of a sharp print would result in a less than optimal result, and you would be far better off just sticking with the sharp print.
One way I could see working to improve image quality while only using the same frame from multiple prints, assuming that the prints were struck from different internegatives, is to track the position of all the grains in the image. For each blob of grain, the center of the grain is recorded along with its luminosity and color. This process is repeated for each of the identical frames. Grains in similar positions with similar color/luminosity are grouped together, and an average is taken of their positions (instead of their color). Finally a new image is created from the data, using the best guess of the position of each grain in the image. Compare this with merely averaging the color of each pixel and you could potentially regain the sharpness present in previous generations of the film. After all, the blurriness is only introduced through grain forming in inaccurate positions on the film with each new printing.
The above method would not work for Technicolor film, however, since it is not a grain-forming copy but rather a copy made by imbibing a piece of celluloid with actual dyes. This process introduces some blurriness of its own, but it’s probably still the best source to use for sharpness since Tech prints of Star Wars were made from an earlier generation of print than the Kodak or Eastman prints. Perhaps the Tech prints could be used in the aforementioned grain positioning algorithm to ensure more accurate results.