Second question: how many data points can a visualisation be before you start to see noticeable lag?
Re: second question, I'm unsure, but I'll ask the team for some stress tests.
With that being said, it has to be done right. Not only that, I'd expect to be 'VR-only', not even just 'VR-first'. A visualisation tool that I'm expecting would have difficulty conveying the same data on a flat screen. Sure, one could argue that you're still only looking into flat screens in the goggles, but the multiple axes of movement allowed by both the head translations as well as hand movements give you superior control vs just mouse input.
If you're looking at two screens with identical images, then that might be a valid argument, but VR headsets provide stereoscopic viewing by presenting different images of the virtual scene from each eye's viewpoint-- this is known as binocular disparity. It's the same principle used in 3D TVs and anything that requires you to wear special glasses.
The "head translations" you're talking about gives the visual system depth cues via motion parallax, where objects in the foreground appear to move faster than those in the background when the head is moved from side-to-side.
These two things together (stereoscopy and motion parallax) yield a very strong sense of "3D depth", called stereopsis. Having controllers with six degrees of freedom (6DOF: translation along and rotation about the x-, y-, and z-axes) to manipulate and interact with 3D data should be superior, as it is no longer necessary to map 2D mouse inputs to 3D operations which would also decrease cognitive load, in theory.