Well, if you dismiss any tensile forces passed to inner longitudinal members you would have essentially two trusses. However, this is essentially what is called a pony truss bridge, and then the vertical members in the trusses and their fixity to the corresponding beams at floor level need to coerce the lateral buckling of the top compression member. So typically the texts (see Galambos' both 4th and 5th editions) care particularly of ensuring the lateral stability of the top chord, and hence of the transversal U frames at panel planes.
But something in your present design, maybe bottom chords underdesigned or the software detecting the U frame intervention on the overall stability and not counted when first sizing the bottom chord is making that your bottom chord is being the critical member. It also may come for the bottom chord members taking loads outside panel points, this might also contribute to them not meeting requirements.
Respect checking, if you want to acknowledge the -shear lag transmitted- tensile stress transmitted if something to logitudinal inner members and as well all the overall effects of lateral stability you better adhere to 3D models. You may directly model some out of straightness according to tolerances and potential vandalism effects and directly then discern if everything is OK.
Since these two effects may be affecting your stresses, I don't see that, except some gross misevaluation of something, maybe detectable just by T=M/z, you can benefit from some 2D analysis, and I see more promising investigating why your 3D model is giving whatg you think unexpected results. Start by looking at the deformation at some big scale, it maybe some required connectivity is not being read as you planned, this is a common cause of unexpected results; then correct the connectivity if such is the case.
