further testing:
without pssm, 600 moving units results in 30-35 fps, depending on camera position (it would be 13-15 fps with pssm):

so I modified the pssm shadow function not to process far shadows, and together with reset/set SHADOW flag of units based on their camera distance resulted in a minor improvement (3-6 fps):

as I realized, the displayed milliseconds do not show exact values when pssm shadows are used. if I calculate the fps from them, it should be much higher. but there are significant differences in polygons per sec, and in the quantitiy of visible entities and triangles. I don't really understand why time_client+time_server values are increased when pssm applied... I should learn a bit of optimization. but meanwhile I started to deal with code clearing/optimization of the unit and group update loops.