Doing what you want effectively means running the backups two times (two jobs, one backing up to the SD at Site A, the other to the SD at Site B) -- A better approach (assuming you back up to files on disK) might be to use something like DRBD (Linux) or GEOM gates (FreeBSD) to duplicate the back-end storage your daemon is using: This provides replication of the data without requiring a second backup job (though it has its own problems if your network link isn't reliable).
Other options include rsyncing your tape-files to an off-site provider (e.g. rsync.net, or your own secondary datacenter if you have one) which provides you with a local and remote copy in the event disaster strikes your main datacenter. The big caveat here is that your "remote" copy is always out of sync by as long as it takes your system to transfer the files over.
For what it's worth, my implementation is a variation on the second method I described: As part of the Bacula mount/unmount scripts the server rsyncs the virtual tape files (on mounting it pulls any changes from the remote site. On unmounting it pushes the changes).
This does make mounting/unmounting tapes take longer, which increases the time it takes a backup to complete (backups hang waiting for the "tape" to unmount while rsync is running), but with sufficient bandwidth between sites (and rsync's intelligent deltas) the overhead is not too bad.