We maintain multiple repositories, many of which have the same rpms repeated in different locations. In order to save space we use hardlinking extensively. The hardlink command does an ok job of finding things to link, but we are only really interested in rpms and don't want repomod.xml or comps getting linked so I wrote a little python script to hard link all the rpms based on a checksum. The script is over at my github. hardlink_rpms
