thanks for the link, but i think i found my favored solution right now, which is extracting the archive into raw files (and header, hashtable, blocktable) and then reassembling the files on demand into a byte-for-byte equal archive on demand (or via virtual file system). this will block align everything, give people access to raw assets and is flexible and performant on the filesystem because of hardlinking. appreciate your help though :)

A virtual filesystem to "simulate" the MPQ files on-demand based off the raw deduplicated assets was the exact thing that came to mind when I read the OP. Happy to help on this. Email in profile.