gurkendoktor

Yes, it does. I have it run with nginx, so I don't care about the .htaccess
Read the instructions on how to get hhvm running here: http://hhvm.com/blog/1817/fastercgi-with-hhvm 
You might need some tweaking and run into a lot of issues - hhvm has not been tested with ProcessWire. I also chose to let it run at port 8999 to be easily able to switch between hhvm and php5 for reasons of comparison. Your mileage may vary, but with a modern system your chances of getting it to work are better.
Some stats using a fresh installation of Processwire with the site-beginner scheme
ab -c10 -n500 http://pwtest.gurkendoktor.de/ php5-fpm: Concurrency Level: 10 Time taken for tests: 23.425 seconds Complete requests: 500 Failed requests: 253 (Connect: 0, Receive: 0, Length: 253, Exceptions: 0) Total transferred: 1902241 bytes HTML transferred: 1700241 bytes Requests per second: 21.34 [#/sec] (mean)   hhvm: Concurrency Level: 10 Time taken for tests: 8.702 seconds Complete requests: 500 Failed requests: 232 (Connect: 0, Receive: 0, Length: 232, Exceptions: 0) Total transferred: 1900304 bytes HTML transferred: 1700304 bytes Requests per second: 57.46 [#/sec] (mean) This is in no way scientific. PHP does not use an opcode cache, which speeds up things a lot. on the other hand, hhvm gets better the "warmer" it gets (ie. the more load it gets).
The nice thing is, that it still works with fcgi_cache:
Concurrency Level: 10 Time taken for tests: 5.110 seconds Complete requests: 500 Failed requests: 491 (Connect: 0, Receive: 0, Length: 491, Exceptions: 0) Total transferred: 1912527 bytes HTML transferred: 1699527 bytes Requests per second: 97.85 [#/sec] (mean) Yes, there are some more "failed requests" - I blame this on ab
Anyway, this is just plain ol' PHP Code. What would be interesting is to actually write a module in hack (the hhvm language) and see if this actually works. Because then you'll be able to harness hhvm's full power and experience an incredible performance boost. I'll figure this out later when there's time. 
More interesting it would be to fork the ProcessWire core and move it to hack. This would incredibly boost the performance, but brings other issues. But I guess this is too much of an effort, I myself wouldn't be able to do it