Indeed. You could do what CRIME does to TLS (I just discovered these attacks).

You could also add a random amount of random padding. It would slow down the attack linearly if the random amount is taken from a uniform distribution.

I wonder if it would be possible to make it slower by taking another distribution.