One scheduler instance, some jobs run twice

From best practices, an average CPU utilization on a back-end server should be within 15%-30% with seldom spikes if needed, but not above 70%-80%. CPU utilization above 90% can lead to all sorts of application anomalies, it's like breathing without an oxygen.

Because your current server is low on resources and very busy, I suggest updating the existing jobs and spreading them several minutes apart so that you don't have multiple jobs starting at the same time. I hope that will solve the issue with duplicate jobs on your end in case it is caused by the job queues not keeping up with the scheduling engine and not having enough resources to timely update job status. There are no other CPUs and the only one is too busy.

Running the scheduler in non-GUI mode may help a bit too as it would reduce the overall system load. GUI mode is much more resource and memory intensive as compared to service and console-mode.