depends on many factors:
- tx size
- do you add to indexes
- do you query indexes during insert
- available memory
- disk performance (for tx-flushes)
- transactions have to be kept in memory (thread local) the use much more concurrency aware datastructures (e.g. CHM) and they must be flushed to disk during commit
I had no issues inserting 3M nodes/s with the batch inserter on my Mac.