As a previous Facebook intern, I was almost assigned to the HipHop team. In one of the internal tech talks, I learnt that Facebook made the transition from running raw PHP to using HipHop around 2009 due to scalability concerns, and it went huge from there:
Wikipedia started to use HipHop in late 2014: http://blog.wikimedia.org/2014/12/29/how-we-made-editing-wikipedia-twice-as-fast/. For a whole list of large users, see here: https://github.com/facebook/hhvm/wiki/Users. Beginning from last summer, Facebook is aiming to port HHVM to more architectures (iirc e.g. ARM).
@kayvonf: Nowadays HipHop is replaced by HHVM, and you might want to update that on the slide. :)
Curious as to the scalability and performance on event-driven server architectures (e.g. Node.js) vs more traditional worker pool servers (e.g. Apache) on large clusters. In the abstract, does one outperform the other on either metric?