Latency to the first byte is one thing, latency to the last byte, quite another. A slow-starting high-throughput connection will bring you the entire payload faster than an instantaneously starting but low-throughput connection. The larger the payload, the more pronounced is the difference.
ehh... latency is an objective term that, for me at least, has always meant something like "how quickly can you turn on a light bulb at the other end of this system"
