Convergence of non-isothermal flow simulation stationary

Hello,

i have problems with my simulation of non-isothermal flow (including surface-to-surface radiation, but i think the problem is the fluid flow). The fluid flow is considered to be laminar. The simulation gives me the required results if i use a time-dependend study and just consider it stationary if the results dont change anymore, but this requires too much computational time.
Why is it not possible (or easy) to use a stationary study here?
Is it a problem with the mesh or the solver?
I tried it with very coarse mesh and very fine mesh, but since its 3D i cannot go much finer without running out of memory...

Any hints are appreciated, thanks in advance!

Edit:
I made a simplified 2D version of the simulation and compared the mesh density. It converges sometimes, but it takes a lot of iterations. And even a normal mesh in 2D is too much in 3D...
Of course i could increase the number of maximum iterations, but it takes too long...
And what does the error message at the coarser mesh mean? (see attached file)