You can detect whether a target is a plain table or not. And use a different path if not.
And independent of using SQL or not: Executing them separately out of process kills you via latency, even if you're on the same machine. The context switches are just too expensive. Which makes single threaded apply untenable for anything with a few more clients, or bulk load.
And independent of using SQL or not: Executing them separately out of process kills you via latency, even if you're on the same machine. The context switches are just too expensive. Which makes single threaded apply untenable for anything with a few more clients, or bulk load.