That's probably the worst possible one.

It delivers pretty much the same result as mine or Eugene's queries, but it creates a hidden loop: it runs a separate query against the source table for each row in the target table.

On tables with some reasonable number of records it will be terribly slow. It's especially dangerous if to consider remote call involvement.

Try the ones from above and compare the execution plans and overall performance.