This compromise is by design for performance purposes; the 'cpu-minutes' value comes from the data for each individual job instance, not the job itself. It's then summed-up to show the total CPU-minutes for the entire job.
The data for each individual instance is only fetched for running jobs (in order to correctly populate the 'Running Instances' view), and when you select a job, which also retrieves the per-frame data for that job. If the per-instance data was retrieved for all jobs all the time, WranglerView would take forever to refresh.
