I think the key (no pun intended) to understand how tieBreakOrder works, is to consider that it's used only with keys that are already (or about to be) present in the tree. That way, it gives a consistent ordering of the keys belonging to nodes in the tree (even when they are not Comparable).

Obviously, tieBreakOrder cannot be used with keys that are not in the tree (e.g. the key you pass as parameter of find), because different instances of the same key will have different identityHashCode, so tieBreakOrder will be meaningless for them.