For the reasons noted by Rick, some Cisco network devices support a "special" responder function (see SLA features), to provide a much more accurate network latency measurement. What this function does is timestamp when the "ping" was received, and when the device finally responds, it "knows" how long it took for the internal response.
Oh, and as pings were not intended for precise measurements, or all requests guaranteed to responded to, sometimes the receiving host, if busy (or it believes it's being pinged too much), just doesn't respond at all. I.e. the ping wasn't lost on the network. Also remember normal "pings" (high) latency and/or loss might happen in either or both directions. Knowing which "direction" a network performance issue is, I recall (?) can also be determined by some SLA tests, but not easily by regular/ordinary ping.
