The job handling has changed during the last years a lot, however, the statistics code is - at least in some parts - still making assumptions based on the old implementation. Therefore the statistics can happen to be useless/inaccurate.
The largest problem with statistics is to find out the number of jobs in the queue. Previously all jobs were always held in memory, therefore this was easy and cheap to calculate. However, only a part of the jobs are now in memory and we have to figure out how we handle this.