After reaching thermodinamic equilibrium, if you wait enough time, any system will reaach, by random chance, any state of lower entropy. Of couse, the time needed to for such a fluctuation to occur increase exponentially with the amount of decrease in entropy (search poincare recurrence theorem).
My guess is that this is not the kind of answer you were looking for. But I have one that you might like. In Wolphran's book a new kind of science he makes a lot of computer experiments with cellular automata. He finds a range of behaviors for the evolution of the entropy for different rules. From systems with standard behavior where the entropy increases until it reaches equilibrium and then fluctuates according to poincare's theorem, to the other extreme in which regardeless of the initial conditions, the entopy increases (but the behavior is trivial in the sense that always converge to a frozen state (usually all ones or all zeros). The interesting rules are the ones in the middle, the dynamics behave in complex ways, but still their entropy sometimes increase and sometimes decreseas, with no apparent arrow of time direction.
Regarding hypercomputing oracles, just my two cents: I believe that for any practical purposes a superturing machine will behave as a random number generator. It will not increase order, at least in the way we currently define it.
Also, assuming a closed non expanding universe) if gravity starts to change (or oscilate) with time quikly enough, then entropy will oscilate with these changes. The reason is that in absence of gravity the universe will flow into a thermodinamic equilibrium with uniform mass density, but in the presence of gravity such a state is actually a state of high entropy (the largest entropy state with gravity would be prety similar to all the mass concentrated at one place.
