This "3d array" is optimized by using two linear arrays (one for pointers, one for data). Once set up, it is accessed using 3 dimensions.
The usual way of implementing a 3D heap allocated array seems to be by making D1xD2xD3 heap allocations, which can result in fragmented memory.
The performance concious way of doing it would be to use a single dimensional array and simulate multi dimensions by calculating, on the fly, how to access the elements. This method is not as nice to use.
The method below allocates a linear data array, and sets up a linear pointer array to give 3 dimensional access to the data array: The elements are accessed in the usual way (eg. pA[1][2][3]). And since there are only two heap allocations, it should be fairly efficient in terms of CPU cache, etc.