The _ResponseBit wasn't set, which made some of protocoltest fail
(ever since it was initially written in 5636aaf so this isn't a regression!),
and which would even lead to a crash (detected by that test) when doing
res->isError() after serializeAndDeserialize which didn't actually create a
Response subclass...

protocoltest went from 3 failures to 1, the last one being the protocol version
(but given the comment next to it, I emailed Dan...)