''the entropy of any closed system not in thermal equilibrium almost always increases''
This doesn't (or at least shoudn't) refer to probability, which is a concept completely foreign to thermodynamics. Thermodynamics assumes (from a statistical mechanics perspective) the thermodynamical limit of infinitely many particles, in which case statistical irregularities are completely absent.
If a system is too small for the thermodynamic limit to be a good approximation, thermodynamics no longer apply, so the question of the validity of the laws for such a system is no longer sensible. The whole thermodynamic formalism and terminlolgy breaks down for such a system, not only the second law. (Just as that the fact that the laws of Germany are not applicable in Austria doesn't mean that they are only almost always valid.)
It also cannot refer to the fact that entropy is constant in equilibrium, since the statement explicitly assumes a nonequilibrium state.
Therefore the statement can sensibly refer only to the fact that there are many systems in nature that are in a metastable state only (see http://en.wikipedia.org/wiki/Metastable_state). Thus they are not in equilibrium but retain their state (and hence don't change their entropy) unless seeded from outside (loss of closure), in which case they suddenly undergo a phase transition.