I was just about to raise a bug against HC4, because the displayed body size does not agree with the size of the downloaded file.
However, I suddenly realised that HC4 is measuring bytes received from the wire, whereas Java and HC3 currently measure the end-user bytes, i.e. after de-chunking. We can just document this behaviour. Or maybe we need to have yet another statistic for HTTP samples which identifies the transport overhead? Might be possible to measure this for HC3.1; I doubt it's possible for the Java implementation.
