You should define indexing policy within your application architectural design.
You may need an on-demand indexing configuration (something like OLTP system). In such systems, you usually add one document per user request. As such, the MaxBufferedDocs option will not affect the system. On the other hand, MaxMergeDocs is really helpful as it allows you to limit maximum script execution time. MergeFactor should be set to a value that keeps balance between the average indexing time (it's also affected by average auto-optimization time) and search performance (index optimization level is dependent on the number of segments).
If you will be primarily performing batch index updates, your configuration should use a MaxBufferedDocs option set to the maximum value supported by the available amount of memory. MaxMergeDocs and MergeFactor have to be set to values reducing auto-optimization involvement as much as possible . Full index optimization should be applied after indexing.
In some configurations, it's more effective to serialize index updates by organizing update requests into a queue and processing several update requests in a single script execution. This reduces index opening overhead, and allows utilizing index document buffering.
 An additional limit is the maximum file handlers supported by the operation system for concurrent open operations