Integrable Hamiltonian systems have the property that, when they are slightly perturbed, their behavior gets only slightly modified. One can state this result in a more technical way through a beautiful theorem due to Kolmogorov, Arnold (see here) and Moser (KAM). In phase space, these systems move on tori and the effect of a small perturbation is to produce a small deformation of these tori. A condition on resonances must hold for KAM theorem to apply. Indeed, for a small set of initial conditions the motion is no more bounded. So, when the perturbation increases, invariant tori get progressively destroyed and chaos sets in. The system goes to occupy a large part of phase space and we are in a condition for ergodicity to be true.
One may ask what can happen when the perturbation becomes increasingly large. A first idea is that ergodicity is maintained and we keep on being in a situation of fully developed chaos. Indeed, this idea is plainly wrong. For an infinitely large perturbation, a dual KAM theorem holds and again we get invariant tori and bounded motion. I proved this in my recent paper (see here). Increasing the perturbation makes tori reform and we lose ergodicity again. Indeed, ergodicity appears to be there only for a limited range of parameters of the Hamiltonian system. This can make us think that this property, that appears to be essential to our understanding of thermodynamics and, more generally, of statistical mechanics, is not ubiquitous.
So, one may ask why all systems appear to behave as if ergodicity holds. The answer to this question is quite straightforward. What makes Hamiltonian systems behave ergodically is the fact that they are composed by a very large number of particles. It is this that provokes the correct working of our statistical approach and produces everyday reality we observe. This conclusion is quite important as makes clear that we do not need ergodicity at a very fundamental level but just at a macroscopic one. This intuition was already present in Boltzmann‘s Stosszahlansatz hypothesis. The existence of a dual KAM theorem makes all this very clear. Indeed, thermodynamic limit can make quantum system quite unstable with respect to coherent evolution producing a classical ergodic system.