The parallel algorithms that Servo uses to implement advanced features of CSS are, as far as I know, novel (block formatting context size speculation, float impaction detection, absolute positioning traversals, parallel automatic table layout, parallel border-collapse).
I vaguely recall them discovering a number of bugs in the spec along the way as well.
Yes, we've discovered several spec bugs, as well as bugs in other browser engines.
Breakthroughs don't always result in research papers. They sometimes end up in products instead. :)
Practical research, not academic research.
There have been academic papers on parallel CSS layout, actually, and my Servo work extends that state-of-the-art. But I've got my hands full with hacking :)
You're the applied physics to CompSci's theoretical physics!