In Oracle RDBMS, if you use multiple threads to insert a very large amount of data that affects a single index segment (i.e., a regular index on a non-partitioned table, or a local index on a partitioned table), the inserts will become slow due to contention on the index. Each thread is competing for latches and locks on the index, which is the root cause of the performance problem. This can be resolved by partitioning the table and running one thread per partition that inserts a subset of the very large data.
Does deleting a very large amount of data have the same problem? I would think that a delete also requires locking a subset of the index, which would block other threads that are inserting/deleting on that subset. However I'm not clear if the degree of locking while performing a delete is similar to the degree of locking while performing an insert. Perhaps the amount of locking is much smaller, or the time required to lock is smaller, and therefore the contention would also be much smaller.
Maybe there are other cases to consider: perhaps deletes block other deletes, but maybe deletes do not block other inserts, because most likely, a parallel delete and insert would not be working on the same data blocks.
Any references would be great.