On the problem of sulfur losses during PIXE analysis