One thing that hasn't been taken into account in any of this "STEP is more accurate than .stl" debate is location. Because .stls are predicated on production, part of the .stl format is location. All valid .stls are translated to be located in the positive coordinate octant. Every user, even if he or she is using the same application as the user sitting next to them, can have their STEP object located anywhere. When the user exports .stl, triangulation can often be different depending on the 3D space location of the .STEP object, ie if you've got a .STEP model of Mendel, and export .stl of two of the frame vertices, it's likely the triangulation of them will be different. If you edit a native .stl with booleans, there's also going to almost inevitably be subdivision along the conjoined boundaries. So STEP might be considered to be "more accurate", but given the built-in inaccuracies of repraps and the outcomes of conversions, none of that's really going to be sustained unless .stl is abandoned for STEP. Given the likely cost and inconvenience of that, and the dramas that'd erupt if anyone forgets to translate the object and their reprap tries to reset all axes to a point on the 33rd floor three kilometres away, I don't think there's a benefit that compels.
