The half model with 4 elements through the thickness of the rectangular tube did not converge to the full (half) load.
I broke the load into 2 steps to use much smaller steps when the previous models got into trouble. Step 1 ramped the force up to 75000 N taking a minimum of 40 substeps, then Step 2 increased the force to the full half load of 85000 N and also took a minimum of 40 substeps. The convergence plot below shows the model getting into difficulty converging.
I stopped the solver after the bisection occurred. This represents 14.6 hours of computation on 12 cores.
The first N-R Residual Force Maximum is located under the compressive tube.
The second N-R Force Residual Maximum is on the Tube.
Since you are applying a force and not a displacement load, I plotted the Force-Displacement graph for a point on the end of the diagonal tube to look for evidence that the slope was approaching zero. It is not.
I believe smaller elements will help this model to converge, however, I don't want to wait 20+ hours for the computation to finish when a 6-element through-thickness mesh is created.
Since all the interesting behavior is at the intersection of the tubes, I recommend you create 4 planes and cut the rectangular tube and the pipes through at a 200 mm distance from the center. Then you can mesh the bodies near the center with six elements through the thickness, while the bodies away from the center can be meshed with two elements through the thickness. Use Bonded Contact to connect the outer bodies to the inner bodies. That way, the solver will take much less time to run.
If you want me to run that model, please make those changes and send me an archive. You might also change the corner radius to get a uniform wall thickness while you are at it.