1 Latency savings are less than request savings because cached pages --- those that can be prefetched --- tended to be smaller than pages that had to be fetched from servers.
2 The individual numbers do not add to 43.1% because some documents are uncacheable for several reasons.