Hugo,
This kind of problem is not uncommon, especially when doing combined airflow/thermal simulations. I have not run into the problem of convergence problems inside of Type56, which is a little surprising. I thought that I had gotten all of the possible TRNSYS errors at least once! The solution I usually use is to implement a short time step (1 minute perhaps) and then to put a delay component (like Type661 or Type150) between Type56's output temperatures and Type98's inputs. It is a bit of a cheat but I think with a short timestep it is realistic enough. We assume that air entering a Type56 zone immediately mixes thermally with the zone air and that thermostats react immediately. When you add the automatic heating/cooling that Type56 can do you are further assuming that the mechanical systems can react instantaneously and provide exactly the right amount of heating and cooling. I don't think it is unreasonable to introduce back a small time delay in order to get the whole thing to converge.
kind regards,
David