For merge joins, one consumer process is assigned for each device in tempdb; if there is only one device in tempdb, two consumer processes are used. The number of producers depends on the number of partitions in the worktable, and the setting for max parallel degree:
If the worktable is not partitioned, one producer process is used.
If the number of consumers plus the number of partitions in the worktable is less than or equal to max parallel degree, one producer process is allocated for each worktable partition.
If the number of consumer processes plus the number of partitions in the worktable is greater than max parallel degree, one producer process is used.