Hmmm, I'd say multidimensional arrays are an extra argument in favour of zero-based indexing. The position of an element in a 2d array is &(a[k0,k1])==a+k0+n0k1 (modulo syntax and ordering convention), where kx is index in direction x and nx is length of array in direction x. For a 3d array it's &(a[k0,k1,k2])==a+k0+n0(k1+n1k2). Using 1-based indexing you'd need to adjust all the kx, which would be an unnecessary waste of space and time.

Edit: Removed asterisks for multiplications, as they got interpreted as markup.