It's not about attacker control, it's about correctness of your code. Without VLAs then static analysis of the call graph (and absence of recursion) is sufficient to prove maximum stack depth; with VLAs then much deeper analysis is needed, if it's possible at all.
