It's just easy to implement constant rate, hence why it's best practice. But adaptive rate might be ok, like, for example, if you only pick from 4 different rates and adapt exactly every second, you only leak 2 bits of information per second. The question then becomes which leaking rate is safe enough in practice?