I am pretty sure that is a typo in the blog post, since the performance of Spark improves as more cores are used, as does Julia's and they both show similar scaling characteristics.

The performance plot would probably be more readable if the y axis was on log scale.

The typo is also now fixed.