The phenomenon you've described - do you observe for all the substrates used for TiO2 films preparation. I think a rational explanation is a substrate-induced crystallization. And as soon as the film reaches a certain thickness - the prevalence of brookite crystalline morphology overtakes the initial anatase.