The small qualifier seemed to be on the size of the documents, and on small documents I see a 2x or often higher improvement. At a scale of 10k docs that's stil

The small qualifier seemed to be on the size of the documents, and on small documents I see a 2x or often higher improvement. At a scale of 10k docs that's still 100ms difference, and in my case can be a huge proportion of the total time spent (more with larger docs, and more again if you include encoding time).