I have been looking a bit for a solution. And I am thinking of this approach:
1. create a python script that traverse the folder structure and processes each image. I think that the geotiff has four gcp:s, each in every corner of the map. By these it would be possible to cut the raster image to get rid of the frame with gdal_translate and in the process compress the files.
Then reproject with gdalwarp to a current standard projection and store all result files in one folder.
2. Let geoserver serve this folder with geotiff files as one layer so that I can display it in my Openlayers application.
That is my best guess so far.. I'm not sure if Geoserver ImageMosaic could handle several thousands of geotiff?
What do you guys/girls think? Is this solution something that could work?
The part that I have most doubts about is the Geoserver-part, my experience with Geoserver (and Java) is not something that give me high hopes when it comes to large amount of data.