Mice could roar but they don’t. There is nothing preventing a small organism from growling; we have horns in India that do it all the time. A roaring mouse is improbable but not impossible. Similarly, a cloud shaped object could be hard, but is unlikely to be so. The relation between clouds and fluffiness is a regularity. Unlike the laws of gravity, regularities are not cast in stone. On the other hand, I am reluctant to admit that regularities are purely statistical in nature, if by statistical, one means relations that aren’t intrinsic to the way the world works. A statistical theory of regularities is agnostic as to the way the world makes the regularity just so; it only cares about representing the likelihood that a given cloud like object is soft. In other words, if I walk around the world punching cloud like objects, I am unlikely to get hurt; but I don’t care whether there is some intrinsic relation between fluffiness and density.

The problem is that any intrinsic relation between fluffiness and density is not physical, or at least mediated by the same physical mechanism all the time; cotton balls and clouds are both fluffy and soft but they are not fluffy for the same physical reason. If at all there is a natural relation between fluffiness and density it lies in the world of embodied information rather than physical mechanisms.

PS: Even probability itself is subject to the same questions about intrinsic versus statistical regularities. Consider a one rupee coin. You toss it a hundred times and it comes heads 48 times and tails 52 times. Is the roughly 1/2 heads, 1/2 tails distribution a regularity or is it purely statistical (whatever that means). The symmetry of the coin argues for a regularity; in other words, a coin comes up heads half the time because it is symmetric and if one can’t control the force with which the coin is tossed, it is going to come heads or tails an equal number of times. In other words, even statistics are derived from intrinsic regularities rather than the other way around.