[dfe]

It's not just the nullability behavior. My experience with several databases is that IN is always (or almost always) executing the subquery then using its results to match the outer predicate. But EXISTS can work the other direction, matching the predicates from the outer query then passing those keys into the exists, allowing use of an FK index on the inner query's table.

[Sesse__]

What databases are those? If you convert to a semijoin, both strategies should be doable.

Note that these caveats do _not_ apply to IN, only NOT IN.