We're starting to do something similar, although without the pairing (we don't do pairing) but with some engineers in the room playing the role of coworkers who

We're starting to do something similar, although without the pairing (we don't do pairing) but with some engineers in the room playing the role of coworkers who are available to discuss the problem, bounce ideas off of, etc. The issue that immediately came up is that within the 2 hour time frame it's just not going to be possible to have a truly realistic block of code from our daily work, there's just way too much domain specific knowledge that one would need to bone up on. So then what happened is that we're back to looking at problems that are in the mold of a whiteboard problem but a bit more complex and with a higher expectation in terms of what their output is. I think this sort of thing works better when a company in the webdev space where companies are using well known frameworks (e.g. RoR) that candidates are going to know how to spin up a reasonable application complete with DB, etc without much effort.