For the (naive) Fibonacci algorithm applied to 20, the following internal run times were measured:
This time, the speed up achieved by indexing is 1.78, and the the speed up achieved by transformation into LL is 3.54 (compared with the indexed version).