Forked from  Issue 722474 

Running expected-longer tests first is a good idea. Easiest way would be if control files were annotated with expected test time, or FAST/MEDIUM/LONG (I think such a control file field already exists but is inconsistently used). I'd avoid anything where we automatically look at job history to compute this, would add another layer of complexity where I don't really want to add one.

Also, note that the benefit of this is reduced if we end up retrying slow tests, as they would end up at the bottom of the queue and be tried again lst.