cpu_per_call not always working properly

We have recently defined a profile with CPU_PER_CALL set to 180000 (1800 sec, i.e 30 min), and assigned this profile to some users.

The aim was to let Oracle automatically cancel long running high cpu consuming BO reports which run on production.  (11.2.0.4, Exadata 6).

The problem is that sometimes we see that the cpu resource limit works well, and query is truly canceled after ~1800 sec (ORA-02393: exceeded call limit on CPU usage),

but somethings Oracle's stops the query only after much longer cpu time usage (7000-9000 sec).

select sql_id,status,SQL_EXEC_START,FIRST_REFRESH_TIME, LAST_REFRESH_TIME,ELAPSED_TIME/1000000 ELAPSED_TIME, CPU_TIME/1000000 CPU_TIME ,error_message