The most straightforward reason is that if you need a transactional database anyway, then moving the queue into the DB allows you to atomically en/dequeue messa

The most straightforward reason is that if you need a transactional database anyway, then moving the queue into the DB allows you to atomically en/dequeue messages at the same time as making other updates. Which can massively simplify your architecture because it eliminates an enormous category of possible failure modes. (Or it can massively improve your system correctness, if you didn't realize those failure modes were possible.)