<app_config>
<app>
<name>milkyway</name>
<gpu_versions>
<gpu_usage>0.25</gpu_usage>
<cpu_usage>0.25</cpu_usage>
</gpu_versions>
</app>
</app_config>
4 wu per amd GPU 0.25
2 wu per NV 0.5 full load on card.
I used to run 7 concurrent on a 7970 just to keep the card maxed as much as possible. The tahiti's chewed through those little work units so fast that a substantial amount of time was spent transitioning the finished WU and loading up the new one. I kept going up until I felt like it was finally using a whole CPU thread which was @ about 7 concurrent, and they all finished in about a minute. I haven't run milkyway in a while but it seems like the credit/day has dropped for most everyone. I was able to squeeze 400k-450k/day out of one 7970.
Anyways, TL;DR. milkyway uses very little CPU, if your cards are going through them fast that the gpu load drops a good amount in transition add another WU in concurrency. If I were you I'd try <cpu_usage>.1 or .2 and maybe add a WU per card. Since you're running 4 at a time and on your 980 at they run for about 1 minute thats every 15 seconds you're dropping gpu usage for probably 5 seconds as it loads up another WU. Try <gpu_usage>.2 and see if you can't squeeze like 10% more credit a day.