Take a discrete time process with mean step size 0. The steps will take values > ε with positive probability.
It seems intuitive that, for any bound, the process will cross the bound infinitely many times. Does anyone know of a result that says this?