I was put through that once, it sucks. You're completely ignoring all the wrong turns you took to get there, the number of people involved etc. Expecting someone to just jump into your shoes and pull of the same trick is not very constructive. What if you framed the problem wrong? Maybe this is the candidate that would have helped you come up with a superior solution given enough freedom and time?

Hmm, this could be a risky approach. If you do it right, it could be a fun, interactive way to suss out a candidate's approach to solving problems, preferred architectures, etc.

However, it also feels quite artificial and stressful, which is a bad combo for a lot of people. Unless the position actually requires rapid fire system design and architecture, I'd be cautious using it as a primary filter. For most positions, designing a system architecture includes include some amount of research, reflection, iteration, etc.

In other words, you're evaluating something the job doesn't actually require (typically), and you're potentially missing out on people who might be brilliant architects given a little time to absorb the problem and think through options.