Currently conformancetest/gentest.html will not add certain expected results to the output, because browser serialization bugs interfere.  For instance, both Gecko and WebKit incorrectly serialize a DOM where the first text node child of a <pre> begins with a newline.  They need to add an extra newline but don't, so when they serialize and unserialize the newline gets stripped and the DOM is different.  The test generator then rejects the test because it generates an unserializable DOM.