Comment by NaiveBayesian

Agreed, that's what I struggle to see as well. It's not really clear why the variance couldn't stay the same or go to infinity instead. Perhaps it does follow from some property of the underlying Gamma/Wishart distributions.