Both XML and RDMSs can be used as datastores, but each implementation has its own advantages and drawbacks.
Using XML to store data for a web application is usually not a big problem until you start dealing with large amounts of data or you decide that you want to discover other information from your data (example: data mining). In other words, storing large amounts of XML files for a datasource is not very scalable but it makes it easier to move the data around. XML can also be used to serialize complex objects in a non-relational format, which could eliminate the need for an ORM if you can serialize/deserialize your objects directly from the XML
RDMSs (databases) are usually more scalable, offer greater concurrency support and are much faster when working with large amounts of data. The relational model makes it easier to data mine later on. Databases do suffer from the object-relational impedance mismatch (http://en.wikipedia.org/wiki/Object-Relational_impedance_mismatch) which might require you to write ugly code or use complex ORMs.