My personal inclination is to believe that all of the ships looked like the 'original' and ignore the inconsistency in the second model, unless there's evidence to the contrary that the original model precludes doing so.
There existed quite a few variants of the model - with different bridge domes, different "impulse crystals", and then eventually the E-B variant. This for me sounds like a sufficient reason to go for multiple variants of the ship, too: at least those represented by the models, and perhaps several more, including but not being limited to some mix-and-match-canon-features ships. That's echoing the real world, both in the sense of echoing the modeling reality and in the sense of imitating real-world diversity of ship design.
I'd also think something called the Great Experiment would be in a state of design flux for quite some time, resulting in alterations of mission profile and onboard gear. Decades upon decades of service would work towards that same result of design evolution.
On the other hand, for a ship whose drive didn't perform as intended, the Excelsior
remarkably retains her warp engines essentially externally unchanged. That to me suggests that the failure wasn't particularly serious... Or that the failed components weren't in the nacelles. What if the ventral cavity, unseen in ST3, was originally to house an all-important transwarp coil? The ship was built around this piece of equipment, and would have required major design changes were the cavity to be removed (the ventral guns were mounted on this structure already, etc.) - so Starfleet left it there, and also built it into the newer ships, even though the transwarp coil was gone by ST4 already.
Thus, the cavity in later ships is doing make-work: it houses a telescoping self-repair crane (which is what the greeblie most looks like) or perhaps a relay buoy dispenser, or then remains free for carrying all sorts of outsize cargo. Usually it's merely exposed to vacuum, though.
There are parallels in the real world: WWI battleships converted to all-new powerplants for WWII, but retaining some superstructures merely for reasons of correctly directing the smoke from the stacks even though the original innards are gone and have been replaced by completely unrelated machines that would better have fit in a completely different structure...