See https://www.w3.org/Bugs/Public/show_bug.cgi?id=19470 for a description of one symptom for this.
See http://xhr.spec.whatwg.org/ for the spec we should follow.
This may sound obvious, but what precisely are we doing wrong? Can you attach a testcase (preferably as HTML)?
Sorry, I was wrong. I thought Gecko did something the specification prescribed earlier (stopping the send() algorithm entirely), but it appears it does not.