the problem with packet loss on the last packets is that tcp/ip only resends packets after a timeout due to lack of ack, or after acknowledging earlier packets multiple times signifying a gap. so if you send 8 packets of data for a typical web page it's way worse if you lose the 8th packet than the 3rd packet. (linux defaults to an initial window size of 10 packets now, windows defaults to 8192 bytes, linux to 14600 bytes. so in linux that would arrive in one round trip time for the data, one one round trip time for the connection setup normally. and in windows three round trips.
but when there's packet loss, linux will normally wait for one or three seconds to resend the unacknowledged last packet. i think there's been a recent change to just resend the last packet extra earlier. maybe with the estimated rtt rather than the timeout value.