As you know the FEM gives approximate solution. Although i haven't done any rigorous experiments but i think reducing the size of elements obviously should give more accuracy. But on the other hand, more elements means more number of interpolation functions (function that defines stresses within the nodes), which may again result in loss of accuracy. So i think making the elements very small doesn't have desired effect. It should be within a reasonable limit so as to transfer the forces to the entire frame or vice-versa. 2x2ft size is quite reasonable. But also don't forget to divide the frames into 2' long segments!
You are right about the connection points of the shells with the frames. I think in SAP2000 version 8 onwards, they have made a provision for fixing this problem. But remember, it is only a mathematical model with geometric and material properties. Although we see it as a 3D section, it is in fact only an abstract grillage of line or plane with no dimentions.
Instead of defining frame sections with predefined shapes you can also define a general frame with the same geometric properties and get the same results. At the joints, only the properties are summed up, not the physical dimensions. So, i think we don't have to worry much about that unless we are interested in the micro details of the individual element.
<<<People also suggested to use a rigid link or small beams to link the plate elements to frame elements. I am thinking to do it this way: set up the girder frames, then set up the plate elements in a layer up from the frame plane where the centroid of the deck is supposed to be. Set nodal constraints (deflection constraints) to bond frame elements nodes to plate nodes above, so that the plate and the girder deflect the same.>>>
Don't mess it up. Keep it simple! I think layer of shells is not the right idea because in that case the frame and the shell will be positioned in two different planes. Remember, these are only the dimensionless lines and planes.
<<<BTW, how does sap do the load distribution if it applies load to girders directly?>>>
Unless there are cross (transverse) girders between the supports the load applied in one girder is not distributed. In that case there must be the deck slab (shell).
Phew, a long post. i really should learn to write in short.