Reported by Francesco Poli on the Debian BTS 6 weeks ago. I thought it was similar to bug#16005, but Francesco confirms that bug#16005 is fixed in mesa_7_0_branch while this one is not.

Here's how to reproduce:
Francesco prepared a test case for another bug (Debian #400406)
[1] http://bugs.debian.org/400406
[2] http://bugs.debian.org/cgi-bin/bugreport.cgi/vtk-multiblockPLOT3D-bug-test.tar.gz?bug=400406;msg=5;att=1

  $ tar xvzf vtk-multiblockPLOT3D-bug-test.tar.gz
  $ ls
  gen_tinyplot3d.f  tiny.q    vtk-multiblockPLOT3D-bug-test.tar.gz
  list_plot3d.py    tiny.xyz
  $ mayavi2

The main window shows up.

 * from menus, select File -> Load data -> PLOT3D file
 * as XYZ Co-ordinate file, select `tiny.xyz'
 * as Q Solution file, select `tiny.q'
 * select PLOT3D:tiny.xyz, tiny.q  in the MayaVi pane
 * in the PLOT3DReader tab (inside the object editor), check "Binary
file", "Do not reduce number of outputs", "Has byte count", and "Multi
grid"; *un*check "Force read", "I blanking", and "Two dimensional
geometry"; select "little_endian" Byte order
 * in the Reader tab, click on the "Update Reader" button
 * from menus, select Visualize -> Modules -> Surface (a blue square
shows up in the TVTK scene)
 * rotate the TVTK scene (by moving the mouse with the left button
pressed)

I should only see at most three cube faces (if perspective theory
is my friend...).
Nonetheless, I can also see the remaining faces, behind the directly
visible ones.
Let me check the opacity settings:

 * in the MayaVi pane, double-click on Surface

Mmmmh, Opacity is set to 1.0.  Hence I should only see directly
visible faces.  Yet, I seem to have Superman X-ray view!  ;-)

Saving the scene to a PNG image, shows that the problem will affect
visualization outputs, as well.
It really seems I'm getting (partial) transparency, even when I do
_not_ want it...  :-(