The reason Indexed rendering is faster than Non-Indexed is due to the post-transform vertex cache.note that indexed will (usually/almost always?) be faster than non-indexed
But you do have to make an effort to traverse the vertex buffer in a way that maximizes the usage of the post-transform vertex cache. It's not automatic and this is exactly what gets confused all the time when people say they are using Indices, ergo it is faster.
If your index list does not reuse the same vertices within next ~20 indices, you're not making use of a post-transform cache anyway.
What was the last time you benchmarked it, anyway ?