I think it refers to the fact that you can't just compute an offset from element 0 to get to an element N, like you could if you'd have an array of structs or classes. Assume a struct uses up 128 bytes, then you can get to element N by using a pointer of N * 128 and you'd be positioned directly at the memory location of that struct. - Edit: cache locality like sibling comment mentions sounds more convincing.