OK, I cracked into the V8 source to take a look at what actually happens. It looks like the implementation does use random access for two-byte strings. However, it also uses multiple multiple string implementations (ASCII, 2 byte strings, "consString" (I presume some kind of Rope), "Sliced Strings" (sounds like a rope again, but might be shared storage of the string contents for immutable strings)), so they could likely use other implementations with whatever properties they choose.
Ugh, I just realized this whole article is a farce. v8 added UTF-16 support earlier this year. Its support for non-BMP code points is now on par with java (although it has many of the same limitations)
See https://github.com/v8/v8/blob/3ff861bbbb62a6c0078e042d8077b2... and https://github.com/v8/v8/blob/3ff861bbbb62a6c0078e042d8077b2....