I was going to correct you, but I'm quite wrong :-)

I presumed that TreeSet reinserts all of the members, in case the comparison
function depends on, say, the default string encoding.  (Say, if you're
going to sort strings by the lexicographic order of their encoded
representation.)  In that case, the order might have changed.  Looking at
the code, though, I see this is incorrect; TreeSet.readObject() assumes that
TreeSet.writeObject() wote the members in the correct order, and it
preserves that order.  So another answer is that serializing and
deserializing a HashSet always creates a correct object, but serializing and
deserializing  a TreeSet may not.