If layout were in JS (or PNaCl), we wouldn't be able to run it off the main thread, or parallelize it in Servo, or experiment with running it on the GPU. We would also lose the ability to do text reformatting on mobile, accessibility would be harder, etc.
To elaborate in regards to running CSS off the main thread: We got very lucky that IE's extensions to allow JavaScript in the middle of CSS evaluation (and also possibly XBL) never caught on, and so CSS can be done more or less entirely independent of the main JS event loop (unless the styling is performed in response to a direct JS query, e.g. getComputedStyle). If layout were performed by JS, there would be no hope: layout could perform arbitrary JS manipulations and we would be no choice other than to stall the main event loop while styling is happening (as current browsers do). But because CSS is largely separated from JS, we can just run it off the main thread and allow JavaScript to run uninterrupted by layout.
Because separating content (HTML) from layout (CSS) from behavior (JS) is a good thing, and because, particularly, being able to have content work without layout or behavior (for situations where you just need the content), and to have content and layout work without behavior (for situations where you just need static presentation) are important.
I'm not saying that its always clear cut or that reasonable people can't disagree on where to draw the line.
Yep. Not sure why they didn't just copy WPF or even Android's way years ago.