If your client is not returning results in time, the feeder eventually wont send the short deadline tasks. Criteria: 48 hours. If returning a science within 48 hours, and is valid, they keep coming. Return results on average later than 48 hours, they stop coming. I.e. the process is fully automated, and thusly it appears your irregular is still regular enough.

I have reduced the buffer/cache settings and this seem to have improved things, but recently I have received a work unit for FAH2 with only 24 hours to compute.
The work unit now har status "No Reply", so I would still like to see this feature implemented, or maybe I should just deselect FAH2.

Reflect on this: If a copy is sent out because of No Reply whilst you computer is still working on the task, and the copy went to a machine also with a half day buffer, it would take 12 hours before the copy gets started. If your original gets reported immediately, and it will because it's being processed in late high priority mode, then a notice is sent to the machine with the copy to not bother (See this often for _1, _2 suffixed tasks), and the task will be deleted, long as it has not been started on the machine with the copy. So technically, there's almost 36 hours (there are exceptions if the backup machine has a larger than > 12 hour buffer depth as these almost always will process AsyncRE in priority i.e. immediately, putting other sciences with longer deadline in waiting to run state).

Even if there's the No Reply, long as a result is returned while it's on the Result Status page, not migrated to master, a duplicate/redundant is also credited **. Then let it be WCG's worry and not yours.

** Not sure if this particular practice also applies to AsyncRE, since for this science results are used to seed the next generation. So, until a member comes bursting in complaining to not having received credit for a late reply, the rule is assumed to apply here too.