With all respect pkh, I don't think thats a very good solution. You will be relying on estimation, which isn't robust.
We need the backup time window as small as possible. So we can't have large gaps between jobs. So with your method if we make the gaps between jobs small, we risk overlapping occuring (and as a result jobs being skipped) as servers fill up with more data over time causing backup windows to increase.
My method of scheduling them all to run at the same time, and then using priorities has worked fine. The first job is overwriting, subsequent jobs appending, and the final job ejecting..