H-theorem and conservation of the Gibbs entropy

[alexV]

My understanding of the Boltzmann's H-theorem is that if a set of a large number of colliding bolls is not in the thermodynamical equilibrium (i.e. the probability distribution function W doesn't obey the Maxwell distribution), its entropy will grow (without supplying heat) until the equilibrium is reached. On the other hand, the Gibbs entropy defined as the integral of W*logW over all phase space is a constant of motion regardless of the system being in thermodynamic equilibrium or not (the latter is a direct consequence of the Liouville equation). How does this two statements reconcile?