Norm has an interesting post about DITASpecialization and Extensibility. I think he brings up some great points.
First of all, DocBook is very customizable in it's very design. What may be missing is a "fallback" mechanism. One thing to consider, though: if you are going to do any useful exchange of content with a partner or between organizations, generalizing that content through a specialization will not preserve the semantic value provided in your customization. In other words, you've invested a lot of analysis and design in creating your customization for a reason; if that customization gets generalized, then your partner/department won't be able to take advantage of that semantic markup.
In order to do any valuable exchange of information, both parties should take a look at the customization and develop appropriate processing expectations. This is where I question the DITA approach. Sure, you will be able to present or process specialized content in a generic manner, but if you want to do useful interchange, then both sides should really meet and develop the processing expectations. If you are going to do that anyways, what else does moving to DITA buy you over DocBook?