While a scheduled task is running, Confluence experiences a momentary timeout when connecting to the database. Because of the brief window of disconnection, Confluence itself is unaffected. However, the scheduled task is not able to complete, and prevents other jobs from starting.
The exact cause may vary, but the underlying problem is that if for some reason the task becomes stuck, the scheduled jobs screen will be oblivious of this stuck state.
Even though the scheduled tasks will continue to be submitted, they will never actually run. The scheduled tasks screen in Confluence Admin will continue to update when the job should have ran, if things were working correctly.
In addition to when a job was submitted, a job should record when it actually finished. This would allow a system administrator to identify a potentially stuck job.
Highlighting consistent failures to complete would also be useful.