The wait type CXPACKET is no biggie. Its the time spend re-assembling the results from parallel queries.
You can change the parameter ‘Max Degree of Parallelism’, to however many cores your big-queries should be shared among. Best practice is the number of core’s in a single numa (max 8).
You can also change the ‘Cost Threshold for Parallelism’ to set the trigger level for big-queries. The default (‘5’) is usually a bit low.