We try to predict the disk requirements for fisheye when we index a 50 GB old subversion server (with hundreds of repositories) containing zillions of commits.
Are there any rules of thumb ?
About 1.5 x the size of subversion repositories on disk. Atlassian's docs are of no great help here, unfortunately: https://confluence.atlassian.com/display/FISHEYE/Supported+platforms
You should definitely watch out for non-standard tag/branch conventions -- if FE is not configured to match them, on-disk size of its cache definitely will be greater.