what I'm saying is that if your developers wrote code to enforce referential integrity - they almost certainly DID IT WRONG.
they did it in a way that would work in a single user database - but not in a multi-user database. And they don't know that they did this.
but unless they used LOCK TABLE or judiciously used select for update to serialize operations - they did it wrong.
how to find it?
select *
from child left outer join parent on (child.fkey = parent.key)
where parent.key is null;
do that against every child table for every foreign key.