It is unnecessarily ineffective.
The source text is contained once in the String field "text", and second time parsed in the TokenHierarchy as a
CharBuffer. Since a CharBuffer over a String doesn't need to copy the content, slight change in the data flow can ensure
the data is shared and not duplicated.
I'll attach a patch for review. I did a few cleanups old code and tried to keep the semantic strictly equivalent, but if
you know the text can be cached in the Text field also in other cases (than ctor), you can simplify it more.

Also, in case that synchronized sections around accessing the text field are there only for proper inter-thread
communication, marking the field volatile instead would make the code more readable, as safe (even more - there's
currently no sync in ctor) and not blocking at all.