I think I’m wrong with suggestion of ‘open air’ type of walls, as I realize now that your mesh is for ‘internal CFD’. Internal CFD means that your walls are not ‘leaking’, as normally a pipe or a tunnel will not leak water or air. So your setup is not common.
If you were using ‘external CFD’, using Zero Gradient walls for top and left / right sides of your tunnel (which will be BMB [Background Mesh Box]) with similar function to your external box now, would make sens.
I do not know if missing thermal results in simulation with 4 custom walls was due to that custom wall setup. 
From previous convergence plots I see that you can shorten the simulation time by 50% without much impact on results. So you are back to mistery of heat accumulation on the top of your box, which in any case looks lake having too big gradients.
Cheers,
Retsam