A couple of points which it may be worth considering.
If ultrasoft pseudopotentials are used, the wavefunctions are not normalised, and the total charge density is not the sum of the individual wavefunctions squared. This is a feature of ultrasofts, not a bug...
The Tru64 version of XCrysDen is less stable than the Linux one (though it was used for most of the examples in these pages).
Back to CASTEP visualisation using free software overview.