I'd love to see where the crossover point is on different architectures...especially when the size of the collection is just over the CPU's L1 cache size. I wonder if the vector's 1-or-many cache miss(es) would exceed the cost of the list's guaranteed-one-or-two cache miss(es).