Additionally, the second "law" of thermodynamics doesn't even make sense at those scales. It pretty much just shows that, in a given (isolated) system, there are way more "unordered" possible states (degenerate) than "ordered" states, and due to basic statistics, in the long run the system will progress towards being in those more probable states.
If the system is simple enough, the probability of jumping to a less-degenerate state is large enough, and eventually, will happen.
Exactly. The probability that shards will meld into a glass pane and water will form a snowman is nonzero, but simply very low. There are no absolutes here.
If the many worlds interpretation of waveform collapse is correct, it's possible that there's actually a universe somewhere in which the entire arctic is covered in such snowmen.
The many worlds interpretation of wavefunction NON-collapse, I think you mean :). And it is a guarantee that if it is not physically impossible for it to have happened, there is a branch of the universal wavefunction as you describe.
I can't find the reference to the giant marshmallow hypothesis, stating that it perfectly probable (although possibly scarcely) that there exists a giant marshmallow that formed somewhere in space out of sheer luck. Probably Terry Pratchett or Douglas Adams.
One might also say that it's a "temporary" perpetual motion machine...