If you have the freedom to experiment, try turning off intra-query parallelism...
SET OPTION PUBLIC.MAX_QUERY_TASKS = '1';
If you want to do some investigation beforehand, have a look for the occasional runaway connection that uses up all the processors by starting a whole bunch of "INT: EXCHANGE" child connections to do some heavy parallel processing.
That may or may not be the problem... but 500 unscheduled requests is an astronomically high number, probable evidence of a nasty bottleneck [ insert Foxhound pitch here :]
(Caveat: High numbers of unscheduled requests are not always cause for alarm in themselves, if the latency is otherwise low and the throughput high... it's like high blood pressure, a warning to investigate further.)
Dedicated servers with steady heavy load from large numbers of busy OLTP connections may not need features like dynamic cache sizing, automatic tuning of the mulitprogramming level and intra-query parallelism... they are primarily designed for changing workloads and database servers that must share a computer with other processes.