Linked list are indeed great for secondary traversal orders (or for subsets).

But I don't understand your comments re in-order CPUs: vectors will be faster there too as they avoid the load-load dependency. Most modern inorder CPUs are still superscalar and fully pipelined.