It's kinda like the uncanny valley effect. The more realistic something is, the more demanding we are about it being realistic.
The overwhelmingly static nature of Daggerfall's environments accentuate this as well. Virtually no 3D objects in the world move of their own volition, and not only would it be less apparent as to what you could interact with if NPCs and items were 3D, but to me it would just make them feel as impassive, uninvolving and gloomy as the world geometry around them.
Not to mention (again) that the finer details on NPCs would look somewhat unusual in voxel form. From a dead-on perspective, or at a distance, yes, a voxel model of the guy in my avatar may look fine. But upon drawing closer you'd start to notice all of the conflicting edges of the voxels and their different shades. It would be very awkward, at least in representing what you'd imagine to be contiguous and seamless details on the character like drapery and hair.