On the Etsy "Code as Craft" blog they've posted an article about their experiences in experimenting with HHVM at Etsy and some of the differences it makes.
In 2014 Etsy's infrastructure group took on a big challenge: scale Etsy's API traffic capacity 20X. We launched many efforts simultaneously to meet the challenge, including a migration to HHVM after it showed a promising increase in throughput. Getting our code to run on HHVM was relatively easy, but we encountered many surprises as we gained confidence in the new architecture.
They start with a brief overview of what HHVM is for those that aren't sure and talk about where their focus was in these experiments. They list out some of the main reasons for trying out HHVM and the role of concurrency in their current application. They started with the "minimum viable product" and compared benchmarks between PHP 5.4 and HHVM on several endpoints. They also show how they "teed" incoming requests to both servers to ensure that the responses were the same across both. They also talk about using employee-only traffic and the overall statistics for when they released the HHVM version internally. They also talk about some of the undocumented features to keep an eye out for if you're thinking of switching: "warming up" the requests to align them in JIT memory, using perf(1) for profiling and the use of the HHVM interactive debugger (hphpd).