Neither MidLifeXis nor ikegami's statements seem to imply a guarantee that an object will be immediately be destroyed when the last reference goes away though.
So even in your second case it would be in theory possible that the objects are freed in a different order than what you're expecting.
So there still two questions that might be interesting: 1) does the spec mandate object destruction as soon as there's no reference left; and 2) do all current implementations (including Activestate) behave the same way?