2D Interpolation vs. Tetrahedral Interpolation
In 2D interpolation you consider four points around the area you want to determine to figure out the value of the color in the middle. You also need to consider the distance between those four points and the farther way the points the less accurate the final value. That is considered a normal interpolation process.

With Tetrahedral Interpolation you take the original four points and sub-divide the area into triangles, thus utilizing three points for more accuracy. When the value you want to determine falls into one of the triangles you are now only using three points in the area and ignoring the fourth point. What is exceptional about applying Tetrahedral interpolation is that when your point lands on the diagonal, the point only references two points between the diagonal and not the third point.

The real issue with determining a a neutral grayscale is that those values always are defined by diagonal. The new apply engine significantly improves how these points are determined. A 2D interpolation between two points depending on the value of the points, means you may be neutral gray at the higher and lower ends of the gradient, but somewhere in the middle you are "less than gray". The output exhibits a "ringing" effect from cyan to magenta in your grayscale gradients. With the Tetrahedral Interpolation you no longer see this issue and the overall processing is faster.

See ICC Build Options.