Re: two-phase send concerns

Maybe the reciever should wait "5 seconds" (i.e., 'an arbitrary time
period') before closing the connection on an abort. That is, if we're
going to stick random delays into our protocol in order to protect
real applications against bugs in the protocol stacks on popular
platforms, we're better off putting those random delays in for the
case where the transaction is being aborted than in the normal course
of events.