I believe what he said, and what I sort of suggested, is that you don't link to the parser directly. A tree isn't really the best way to have all the information about your game. There are more efficient ways of accessing information.
Also, making your engine depend on JDOM primitives might not be a good idea. If some day you want to change JDOM to something else, all you need to do is rewrite the one function that gets all the data from the XML tree, and puts it in the according datastructures, within your classes.
If you have JDOM primitives (nodes, I'm assuming) scattered all over your classes, it'll be a pain.
So, I guess the suggestion is: use JDOM as an interface to get all the hard data out of the XML. Copy that data over to your classes, and dump the JDOM XML tree.