Not a concept, just an idea :

Making more use of duration_correction_factor in the host table could help to adjust the rsc_fpops_est values of generated workunits.

In the first step, BOINC should supply some SQLs that help the project people adjust their estimates.


Later the WU generator itself should be able to query the current average correction factors of a WU series (only active hosts!) in order to (slowly) adjust that rsc_fpops_est.

Some asymptotic behaviour similar to the calculation of the duration_correction_factor would fit there :
new = (old * 9 + current) / 10

Adjustment needs to be done by WU type as some projects have different correction factors for their series of WUs.