I have mixed feelings about this sort of thing.

As an interviewer I've done something like this when needing to hire for a very straightforward job with very little time and an effectively endless supply of candidates.

This addressed the problem of our time crunch, but I doubt it helped the selection process overall.

I think this sort of testing is subject to all the same issues as most any other method of interviewing, but limited to a much smaller sample size.

Basically, it's arguably good for the interviewers and a coin flip for everyone else.

As an interviewee, I haven't encountered a test like this, but I expect my response would almost surely involve questioning the architecture decisions that led to such a problem in the first place.