As someone else said, if the VM is NUMA aware, make the MAXDOP no more than the PHYSICAL cores in each NUMA node. For other configs (in any case actually) to give best advice I would need more information about host/guest config, workload, etc.

Cost threshold for parallelism default of 5 is universally too low. 20 for OLTP and 50 for OLAP without other information (which as a performance tuning consultant I will always have).