Gradients are a real pain when making mosaics from a light polluted site.
By reprocessing the final stack separate R, G, and B fits of some data from mid December 2012 with StarTools the previous obvious gradients are now gone. I just used auto stretch, contrast and noise reduction in ST. The step that eliminates the gradient is the contrast module.
The join between the two panels is now undetectable. The dim or faint detail is also far better. It could be almost as good as a dark site. The image is the final evidence.
Here is a large image of a two panel mosaic at the sensor native pixel size. 10MB
http://d1355990.i49.quadrahosting.co..._05/LMCmos.jpg
Bert