A lot can be said for having an engineering team with the experience and maturity to recognize both strengths and weaknesses of their technology choices.
They wrote a blog post about it but it's been a few years since I read it. IIRC it centered on the inherently stateless request model and how that interacts with the way they wanted to do concurrency.
Most people recoil from it but their reasons were sound and they played well off its strengths and planned for and mitigated its weaknesses.