Generally, the finer the mesh, the more accurate your results will be.
A good way to check if your mesh is 'close enough' is to add a Error plot. (right click on results->Define Stress Plot->ERR) It will tell you how much the stress is changing from node to node (as a percentage). Large changes in stress will generally lead to errors.
For static analysises, a good way to lower the error is to run h-adaptive loops. Right click on the top of the study's tree ->properties->Adaptive Tab. Activate h-adaptive loops. After your study solves, the program will look at the areas where the stress changes rapidly, automatically remesh those areas with a tighter mesh, and resolve.
Hope that helps
the force balance should be pretty good regardless of mesh but you may see some difference between the 2 if your mesh is really coarse.
what's your thought on this related to submodeling?
It depends on how strictly you define "affect." Yes, to a point there is a change but it will be miniscle. Changes in the reaction forces will be determined by changes in the deformed shape. The size of the mesh will have little effect on this.