Hello,
Have you checked that the same number of core orbitals has been frozen in Gaussian and Turbomole calculations? The default "freezing" scheme in Gaussian is a bit different from the scheme used by define. Furthermore, did you really run RI-MP2 calculations in Gaussian03? RI-MP2 calcualtions with Gaussian03 should not be possible, at least according to the manual (the density fitting scheme in Gaussian03 can only be used for pure DFT, and according to my small test calculations their scheme differs from the RI-DFT used in Turbomole).
Regards,
Antti