I haven't really looked into this as I am working on projects in other software.
My assumption is that by just deleting the text in the ID document, ID is not removing the nodes in the imported XML and ID is simply reusing the imported XML to export with. Which is what facilitates round-tripping content.
You might experiment on a copy of the publication--and likely backup the document and source files, too. With the copy open, try deleting the node as well.
That said, I typically use XSLTs to alter nodes/information in the XML either pre-import, post-import, or both. Sometimes I'll simply write some javascript to run in my programming text editor and do it there. Which ever is easiest for the task at hand.
Mike