Calculations can be performed using this geometry, but optionally it can be refined further, using the RUNTYPE SADDLE directive in the input file. (This corresponds to the TS algorithm in MOPAC, not the MOPAC SADDLE mentioned in paragraph 2C.)
Actually, the Diels-Alder reaction is not a good example for this type of calculations. It has been shown that elaborate methods have to be used to reproduce the synchronous mechanism. It is likely that a simple SADDLE calculation in Gamess, using the MOPAC TS geometry, will fail to optimize at a 'synchronous' saddle point. So this part is likely to change in the future.
However, AM1 calculations have been used to explain regioselectivity in Diels-Alder reactions. For a 'modern' virtual reality approach, see a contribution by Suñer, Casher and Rzepa to the ECTOC.