Presently the clean method of the AbstractIpAddress class only validates against IPs existing in it's own subnet.
As per discussion in #114 an IP existing in a parent or child subnet should be considered used.

If this is valid I've already started on a patch, working up the tree is straight forward.