Smaller work units?

I'm fairly confident your problem is to do with your machine, how it is configured and the preference settings you are using, rather than anything to do with task size. You seem to be mainly doing Arecibo BRP4 tasks (just one BRP5 (PAS) task visible) and the crunch times are totally out of kilter with what should be possible. You appear to be using only the GPU (HD7770 Cape Verde - or is it a 7750?) here so I guess you are using the CPUs at other projects. Do you have CPU tasks running on all CPU cores while a GPU task is running? if you can optimise your setup, you should be able to do several BRP4 tasks each and every 18 hour period.

As an example of what is possible with a HD7770 (Cape Verde) GPU, take a look at this host which has a similar GPU. It's powered by a 4 year old Phenom II X4 but if you drill down into the tasks list you will see that BRP5 tasks (which take more than 3 times longer than BRP4) are completing in about 8Ksecs each. The machine runs 3 GPU tasks in parallel and 3 will complete in about 24Ksecs. So your machine taking 85Ksecs for the much shorter BRP4 tasks is a real problem that you need to sort out. Perhaps you'd like to give us a lot more information on your setup and preference choices? The most likely culprit for the poor performance may be the PCIe slot in which the card is inserted. Is it a PCIe1 or PCIe2 slot? Is the slot running at 16x speed? What is the model of your motherboard? Is it DDR2 or DDR3 RAM?