It's not only that, it's also that if you have two languages in your codebase you now need two ways to deploy, two types of application servers, two types of testing frameworks/QA setups etc. If having the two languages means you can create a product only marginally faster/better then it is not worth all of that overhead. As mentioned in the article, there are places where the cost becomes worth it, for instance faceted searching is done in Java via Solr at Etsy. But for the most part fitting your problem into the existing infrastructure is a lot better for the organization than bringing in the perfect technology.
