I'm genuinely shocked that this made it to the top of hackernews. I would expect better. While the idea of having the Python interpreter in the browser is very likely better than having a compiler Python -> JS (if we care about Python), the implementation here is both half assed and really bad. It takes quite a bit of effort to create a python interpreter that can be called python. This is nowhere near there. On top…
Python doesn't specify the complexity of dicts, that they're "efficient" is a CPython implementation detail other runtimes happen to copy. Even a worse-than-linear time mapping that satisfies the protocol is still a 'dict'. Surprised you don't know this given your relationship to PyPy.
As for whether I'd prefer a 100kb "not quite Python" or sending a few MB of JS (that takes 10 minutes to translate for each change) to the browser just to handle a few functions in a cross-platform manner, I'm glad to see more options when making this choice.
I'm in the midst of evaluating a solution to running simple mathematical functions on a Python backend or in-browser (for graphing). In a case like this, a handful of crappy guarantees about the lowest common denominator arithmetic is all I need in a solution. It's easy to get wrapped up in correctness, but ease and practicality trumps it almost every time.