I'm not sure how many levels it was applied but my general strategy on it was the first obfuscation method seemed to be an IIFE that triggers eval on a big string that has had some transformations applied which then calls eval on the next level and so on so I would take the IIFE, turn it into a function definition stuck in a variable like "decodeFunction1" so I could just recursively call directly and change the ending to return the string instead of eval it directly. There is probably something that could be done with debugger breakpoints here but I don't know enough about how that plays out inside eval. Anyways if I were evil I'd make sure at some point in this chain there is a subtle change in what's happening that breaks this approach so I wouldn't be surprised if someone told me that was the case :).
It'd be interesting to see what some real JS devs could get to, both from a "how hard is it" perspective but also just to see what different obfuscations are used in the real world.