First, should MATLAB use all of the CPU capability? That is difficult in what you are doing, becaue it is not easily parallelized. The scatteredInterpolant is doing its work using a 3-d tessellation. Each point will lie in one simplex of the tessellation. So it needs to decide where a point lies, then interpolate inside that simplex. What happens is this is not necessarily easy to do in a way that uses all of your cores.
Some things in MATLAB that use linear algebra or some other operations are automatically parallelized. So it can be the case that you will see all of your cores running flat out. (Yes, I checked the other day on something that I was doing, and it was nice to see MATLAB running at 800%, thus all 8 cores on my machine, flat out. But then the easy part finished, and it dropped down to 100% of only one core, since the next step was not amenable to multi-core processing.) Perhaps a future release will see TMW include multi-thread processing in scatteredInterpolant, but that seems to be not the case today. They have only so many resources to apply to such enhancements.
(A quick check shows that scatteredInterpolant seems to use only one core on my 8 core hardware, even on a large problem. Done in R2019a.)
Next is the question of whether you are even doing this the correct way.
It seems from your question that you have a regular grid in the (x,y) plane already. So all points lie on that grid. But then you have points in z for each (x,y) pair, that are NOT regular. Your goal is to interpolate these points onto a 3-d lattice in (x,y,z), and the way you are doing it is to just throw it into scatteredInterpolant.
Effectively, it seems you are using a 3-d interpolation in a way that really needs only a 1-d interpolant.
My suggestion is to just reduce this into a 1-D interpolation, although many of them. At any (x,y) location, interpolate along z to grid the result. Each such interpolation is trivial and fast. It will run like blazes, even in a loop over x and y.
Don't use a high horsepower tool to solve a simple problem.