[tuxor]

This pdf consists of pictures with multiple layers. One layer for the text (~800KB per page if saved directly from Evince as png or jpg in decent quality), one for the background (and two additional layers I don't understand) - really good work has been done here in extracting the text layer after scanning - I wonder whether you could even remove the yellowish background layer without losing any quality of reading/information? Then of course you have the plain text information from OCR which doesn't amount to a significant part of the file size.

By the way, showing this pdf in Evince is really slow on my notebook with 2.4 GHz Core2Duo with 4GB RAM - so I'm not surprised it's kind of a challenge for the M92. Printing it with cups-pdf is slow and returns a file of appx 1GB (only printed the first 20 pages for testing) that doesn't really contain what you'd expect.

Extracting all images with command "pdfimages" yields 3 ppm files and 1 pbm file (image format with only two different colors) per page. All images together amount to more than 8GB (I estimate). If you only keep the pbm files, which contain the text information in appx 2000x3000 pixels, it's about 320 MB. Convert those pbm files to png and you have appx 30 KB per page, so all in all 30*400=12000KB=12MB for the text layer in the whole PDF extracted as PNG.