it is not too far away from approach introduced by javadoc / doxygen beside the datamodel used behind this approach has much more value for additional information integration as provided by the classical approach.
not all yet done by the "out-of-the box" solution provided here: http://www.ibm.com/developerworks/xml/library/x-DITAdoclet/
but you can imagine the possibilities using this kind of approach:
- integration of additional information provided by marketing or tech-doc groups or service groups without any media break in information usage
- usage of other deployment processes for api documentation available within companies infrastructure