Hello,
What changes when the mesh size changes is the Courant number, and this is a hint. You could have inf-sup oscillations, in this case the free surface gradient compatibility should have an influence, try 0 instead of 0.9 just to see.
Another thing is that you have no diffusion of velocities, try 1. instead of 1.E-6.
I do not believe in a parallelism problem, you can check that you get the same problem in scalar mode.
If nothing works also try :
TRAITEMENT DES HAUTEURS NEGATIVES : 1
OPTION DE SUPG : 0 ; 2 ; 0 ; 0
if you do not mind having slightly negative depths in dry zones, it does not spoil mass conservation and is faster.
With best regards,
Jean-Michel Hervouet