That result is actually why I am healthily skeptical of the performance comparisons specifically in this post. Like the previous comment I too have been in a situation where JSON encode/decode on large payloads were significant enough that we decided we needed to figure out something more optimal. Our solution was to use ujson instead (of the built in) and that was good enough that we moved on.

When I say skeptical what I mean is that I have no doubt the measurements are accurate (as in not fabricated); however, I wonder what the full context of the benchmarks are and whether or not it's representative of situations like mine where the built-in parser started to fall over so we replaced it. It may also be the case that the PyPy parser would also beat ujson in this case, but if we are limiting our test case to just those where the CPython implementation is marginally faster than ujson then I'm not sure it's that useful a benchmark.

I too cannot comment on the quality of the implementation or the relative performance against other parsers...but it solved a performance problem for at least one project of mine in the past.