Summing a vector vs. walking a scattered linked list

Two traversals process the same total amount of data. Traversal A sums a large std::vector<int64_t> linearly (for (i) sum += v[i];). Traversal B sums an equally large std::list<int64_t> (a doubly linked list whose nodes were allocated independently over time and are scattered essentially randomly across the heap), following node = node->next each step. Both are far larger than any cache level. Why is B typically dramatically slower per element than A, well beyond what the extra pointer indirection alone would predict?

17 of 47 in Computer Architecture