We calculated electronic properties of a material system by CASTEP program using different pseudopotentials and find nontrivial difference in total DOS values that looks very unusual to us.
Apparently the pseudopotential, don't affect the the electrical properties which are controlled by outer shell valence electrons. However, the shape of the pseudopotential depends on the choice of its empirical parameter (form and structure factors) do affect the DOS, specially for deep energy bands, which are close to the nucleolus.
Provided you use good, modern pseudopotentials using the same exchange-correlation functional, then there should not be any significant differences in the valence states when you change pseudopotential, aside from a rigid shift in all the bands. If one of the pseudopotentials treats shallow core states as valence and the other doesn't, then of course there will be some differences there, but otherwise they should generally give the same results.
The only real exception to this is if you're using a Hubbard U, because CASTEP applies the Hubbard potential on the pseudo-atomic states and these are different between different pseudopotentials.