SCENARIO
1. You're using MySQL 4.1x with foreign keys.
2. You have table t2 linked to table t1 by a CASCADE ON DELETE foreign key.
3. t2 has a UNIQUE key so that duplicate records are unacceptable.
3. You have a REPLACE query on t1 followed by an INSERT query on t2 and expect the second query to fail if there's an attempted insert of a duplicate record.
PROBLEM
You notice that the second query is not failing as you had expected even though the record being inserted is an exact duplicate of a record previously inserted.
CAUSE
When the first query (the REPLACE query) deletes a record from t1 in the first stage of the REPLACE operation, it cascades the delete to the record that would be duplicated in t2. The second query then does not fail because the "duplicate" record is no longer a duplicate, as the original one has just been deleted.