The problem, I think, is when people conflate coordinates and properties. Each dimension is a component of the coordinate and is expected to increase the extent of simultaneously indexable positions, each of which is assigned its own properties configuration.
There is a common conceit to treat a vector of possible properties as one (small) dimension. For example, we can make your light field hold an RGB value at each coordinate. We can alternatively interpret this as a 5D field of 3D luminance vectors or a 6D field of luminance scalars.
However, if someone writes out a set of these light-field measurements as individual measurements, they might append the coordinate vector and the properties vector into one longer point (x, y, z, a1, a2, r, g, b) and claim it is 8D. However, that perspective ignores that a real light field cannot store more than one color combination at a given (x, y, z, a1, a2) coordinate. An arbitrary 8D point cloud can represent nonsensical, sparse superpositions of multiple light fields.