I have not seen benchmarks for this but I would imagine that a merge would not be much slower than a pure insert because the merge would quickly identify that all rows would need to be inserted.