G98 CASSCF Efficiency Considerations

Dear listmembers,
 I am trying to perform with Gaussian 98 a CASSCF calulation with 4
 electrons in 4 orbitals for a diatomic molecule using a basis set of 224
 functions. I have encountered many problems so far regarding efficiency
 and I would like to ask if anybody has any experience with this.
 First of all, I am unable to do a "conventional" (SCF=Conven)
 because I have g functions in the basis set, whereas the subroutine
 performing "conventional" calculations suppotrs only up to f
 The direct calculation of integrals needs much memory which is not
 available. In fact I get a message:
   InCore calculation needs more space: 341070520
   In this calculation the integrals will be
  calculated in every iteration
   and Contracted with The Density Matrices
 so the actual ("direct") calculation needs very much time to be done.
 Furthermore, the CASSCF iterations after approaching convergence (very
 very slowly), they start to diverge (very very slowly again), until
 reaching the max. no of iterations (64 in this case) and then stop.
 I tried to change the convergence options by imposing quadratic
 convergence ( (CASSCF,4,4,QC,FullDiag) but no luck, I get an error message
 at the beginning of the iterations) or Davidson Diagonalization (same
 convergence behaviour as mentioned above).
 I also use a very good guess (I have compared it with Molpro which takes
 so much less time and converges easily).
 Does anybody have any experience with this?
 Best regards,