Gaussian03: MP2 before CI or CC

 Dear all,
When you run a CI or CC job with g03, it starts by resolving the HF equations. In the next step, the integrals are expressed over MOs and then the first-order MP wavefunction (and the second-order MP energy) are computed, all of this done in the link 804. Finally, g03 uses these informations for the CI or CC job in the link 913. I would like to find references about this procedure. I guess it is related to the Davidson scheme for diagonalizing the CI matrix. But why always using the MP wavefunction instead of the HF one ? Is there an option in g03 for using the HF wavefunction ?
What is more surprising is that the CI or CC energy depends on the MO eigenvalues ! I just slightly modified the code in link 801 for changing one eigenvalue (rwf 547) and the final CI or CC energy varies with this eigenvalue ! I checked the same behavior on different machines. Can someone explain what is happening ? I guess I am missing something but I cannot find what. 
 --
  Dr. Nicolas Ferre'
  Laboratoire de Chimie Theorique et de Modelisation Moleculaire
  UMR 6517 - CNRS Universite' de Provence
  Case 521 - Faculte' de Saint-Jerome
  Av. Esc. Normandie Niemen
  13397 MARSEILLE Cedex 20 (FRANCE)
  Tel : (+33)4.91.28.27.33              Fax : (+33)4.91.28.87.58
  Please avoid sending me Word or PowerPoint attachments.
  See http://www.gnu.org/philosophy/no-word-attachments.html