One nice—and, I think, underappreciated—thing about Servo layout is that the parallelism forces you to organize your algorithms cleanly. Unlike every other engine, render objects in Servo are not responsible for laying out their children recursively; the higher-level parallel traversal driver does that. That means that you must write your layout code in such a way that the right information is available at the time the traversal invokes your layout method. You are also limited in what you can access: your render object can't access your parent (because that would be racy), nor can it access the DOM (because we can run layout off the main thread). This requires a fair bit of up-front thought, but once that's finished it's easy to read the resulting code, because the layout code is grouped into specific functions ("assign-inline-sizes", "assign-block-sizes", and so forth) that do just one thing. This goes a long way toward making layout easier to understand, especially when it comes to complex situations like tables.

Right — and it's hardly surprising that Servo's codebase is in many ways nicer than existing browsers (both because of the inherent separation parallelization causes, and the lessons learnt from existing browsers). And given layout has always been arguably the worst part, it's hardly surprising that Servo shows good gains there.