I suspect that the performance advantage of allowing the CPU to cache-prefetch many pointers at once (vs. having to follow a pointer to get the next pointer, in the case of the linked list) still makes a vector of pointers better than a linked list for sequential traversal.