I saw the video, and read the docs. It seems like the backend is basically an enfilade structure (in the xanadu sense -- i.e., it's a god damn nth-order b-tree with splitting on demand rather than based on numbers). I like enfilades (and a lot of the other xanadu data structures) so I'm happy with this -- I'd be happier if they used enfilades the way they were meant to be used (with tumbler addressing and transfinite numbers, because it's FAST) but google is primarily a web service company, not a systems programming company, so I can understand that they might ignore the most bloatsaving features of a pretty nuanced idea (it took me several years to really understand the draw of using tumbler addressing and enfilades). Ted Nelson did a google tech talk, and Plan R has been in effect at xanadu since around 1993, so presumably there is an implicit goahead on using this (potentially nasty, licensing-wise) data structure.
The thing is, the idea isn't new -- it dates from circa 1961. This makes it older than two-dimensional arrays. Google has re-branded it, sure. They have put it to use. But sooner or later someone is going to realize that a competent coder can whip this up in a couple of lines of python (the back end at least -- and if third party apps are just chunks of javascript and sgml in the istream, the frontend won't need to be much more than parsing through and spitting chunks into the vstream). In fact, a competent coder could pretty easily build a majorly improve (on a fundamental level) system based on the same design -- for instance, a couple weeks ago a friend and I were walking around town and managed to work out the design for a transparently distributed enfilade with redundancy, based around a normal SQL database and a modicum of clever message passing. This would be significantly easier to implement in, say, erlang, than a markov chain or a brainfuck interpreter. Granted, both of us have been working with various forms of the xanadu code and documentation for years, but once you get past some of the special terminology the ideas aren't that complex.
I applaud google for putting much-needed tech out there, especially when it has been kicking around for fifty years. I just think they are facing a situation where they can either do a half-assed implementation or put themselves shit out of business (or at least up the creek).