Did you notice the variation in names of the workunits? There are two variants: WU-names with Rigid and WU-names without Rigid. The ones with Rigid run much longer than the ones without. The short ones can take an estimate of (much) less than 10 minutes and — when they're well on their merry way — the long ones will catch up sometimes, now and then. What you see is that the runtimes are interfering with each other, that's why you see some sort of 'mix' of the estimates, indeed. That said, the estimates will keep varying wildly because the runtimes vary wildly and they will keep on doing that. I'm afraid you should try to get used to it, unless someone finds a solution.

I looked at the task properties last week and found the task sizes (in GFLOPs) was the same across all of the tasks. Looked again this morning and the RIGIDs are ~4X now. So this may be getting better. However, I now have ~60 RIGIDs in my queue that are set to timing of smaller tasks so it will take days to crunch through them. Estimated time will not adjust until new tasks are downloaded (from what I have seen). Just trying to get my FAHV 10 yr completed. As long as the tasks keep coming for another month it will work.