I definitely don't know all Nastran's eccentricities, but I guess starting off, the NX/MSC CQUAD4-33 (PSHELL centroidal) and the NX/MSC CQUAD4-144 (
bilinear) use the same formulation. It's just a different way of writing the data. The CTRIA3-74 has a constant strain, so there's only 1 result per element. There are two sets of higher order elements (CTRIA6/R, CQUAD8/R and CQUAD/CQUAD9). For years I only knew about the first set and those have N//2 + 1 results (so 3+1 for a CTRIA6/CTRIAR). I believe the second set is related to NX solution 400 and those also have only 1 output location. That's about the point you start seeing differences between NX and MSC in terms of element numbers.
As to your question, yes it is by design that there are two classes for the same type element (PSHELL). If the full results were recoverable, then I would have no problem with it being less transparent, but you can only go one way because you don't have the gauss point results. Basically, you need to understand your data in order to get the right answer.
NX recently added a new solid output format that doesn't include the centroid, 9 principal direction components (3 vectors) and I believe also leave off mid/max/mid principal stresses/strains. As far as I can tell that's a
runtime optimization and I'm on board with it. Those classes can convert between each other.