Have you measured _transactions_ per second? For both “insert only” and “mixed insert/update/delete?” This might be a more telling metric in terms of the stress it puts on the system. There’s a big difference between 1 transaction with 150k inserts, and 150k transactions with 1 insert each – especially when the workload shifts to include updates in the middle of those.

If you’re doing insert only, you might benefit from copying directly to disk rather than going through the typical query interface.