I am not 100% certain of this, but from observational data, it seems to me that the rate at which your BE lowers once you have drafted more peasants that you need to fill all your jobs is quicker than the rate at which BE recovers once you have more peasants than you need to fill all the jobs.

Is this deliberate, or just a quirk of the code that could be examined and adjusted?
If it can be adjusted, the rate at which BE recovers should be identical to the rate at which BE lowers due to population variance.