Certainly his famous XML rant was informative and provocative. The thing that still rings in my mind is the argument that presentation and content are not really separable. Still chewing on that one.
There were many minor bits of analysis that he did to code, taking up challenges about performance, for example. I remember one example where he clearly spent hours on one performance claim, rewrote the proposed code within an inch of its life and gave an irrefutable argument about what the performance was.
His deep knowledge of SGML (I suspect deeper than all but a handful of folks) was a great source of insight on comp.text.sgml for what seems like a long time. His many rants about C++ were informative as well as confrontational. He told of working in C++ and generating very nice looking code, to the satisfaction of the client. He cheated a bit, however, as he used Lisp to generate the code.
There have always been reactions to his blunt, confrontational style. There are subjects, however, for which there are a right and a wrong. Erik wouldn't hesitate. Did not Richard Feynman reduce an ill-prepared presenter to tears with relentless questions?
Good teachers are not necessarily putting the student's feeling of self-worth first.