Among all GLS estimators, Why is the GLS estimator with the trueachieves the smallest variance?
It's a consequence of the Gauss-Markov theorem. You just transform the GLS problem by left-multiplying by \Sigma^{1/2} and then apply Gauss-Markov.