The mesh can be improved. There is a volume within the main pipe, downstream of inlet 2, that has a high velocity gradient. However the mesh size is large at this location due to the automatic meshing process. See the post-processor screen shot Fine (U - cell data).
(I've had some difficulty pasting this image into the comments - its tiny!. Any tips would be welcome).
My next step would be to manually generate the mesh with a mesh refinement region. And then carry out a further mesh convergence exercise.