In order to optimize the throughput you have to distinct between different operations:
Content creation and updates are single item operations, which you can run concurrently in order to increase throughput.
Practical experience has shown 10-15 concurrent requests to be an upper limit after which scaling is greatly reduced. Also keep in mind not to overload the system if it is used for productive work.
Bulk repository operations, that is publish, unpublish, and delete, don’t profit similarly from concurrent invocations. The first phase of a bulk operation, like setting the content approval flags, can run concurrently with other bulk operations, but the final publications are executed strictly sequential.
For bulk operations you can influence the throughput with the size of the passed-in content set. The recommended range lies between 100 and 200 content items. Larger sets do not improve the speed anymore but block the publisher for longer. They can also improve the risk of network problems for synchronous calls (which are not recommended, see Bulk operations).