I do not understand the on_frac figure. Mine is .860740. I checked it last night and again this morning. After leaving the computer on all night it was the same. If it is the fraction of on time divided by elapsed time than it should have changed at least in the last digit. There has been less than 50,000 hours since I installed boinc on this computer so since the figure goes to 6 digits the last digit is still under a hour in time so 8 hours of on time would definitely effect it. For instance if I go to results status and check the cpu time/elapsed time, I see that the elapsed time is always greater than the cpu time even though the computer is on 24 hours a day. The worst one was 94.89/108.57. That would mean in those 108.57 hours this computer was on only 94.89 hour(about 87%) but this computer has been on almost 100% of that time.

In amongst, if you run cep2 on hardware that is not suitable to run these heavy model and i/o applications, you're also going to see poor efficiency. Typically mcm, faah and fahv run > 99 percent efficient on my pc. Nothing else running, really nothing but passive incoming file scanning by an av program. Even a sole cep2 makes > 98 percent, but if multiple are run it quickly drops to below 95.

Lol, but the correct is as what i wrote 'Elapsed is the time the agent is allowed to compute.' which can still be less than when the job started to when it finished. The simplest case, task switching the agent may undertake per the scheduler. The result log could give you a much further apart 'run' time than the actual 'allowed to run' aka elapsed time.

There is no such state as valid pending. There are lots more results pending which do not show in the results section, I went from 2 pages to 5 pages, that explains why there are less results/cpu time.

Not sure what these cpu times refer too, but it seems simple to me. The weekend! People turn off their computers/offices are closed.
In the case of IBM and the banks, this is a huge amount of time that is "lost".