Like all interview questions, if they are not structured and scored, they're unreliable as predictive tools. I'm always amazed that data-driven companies ignore reliability and validity when hiring and evaluating performance.
His reasoning for using puzzle questions may have been face valid, but there's no evidence that they're related to resilience (or anything else they're supposed to measure) and the little research that has been done has shown only a moderate relationship with cognitive ability(http://www.researchgate.net/profile/Chris_Sablynski/publicat...).
I agree that a trial would be best if not for the lack of time. A good alternative is an assessment center, which is a structured simulation of the job. They're difficult and costly to develop but once they're complete they can be reused year to year.
But the relationship with future performance and these subjective "reading between the lines" opinions is unclear. You need standardization to reduce bias in hiring decisions and make them as objective as possible.
Industrial-Organizational Psychologists have been struggling to change this for decades, but we're a small group and the scientist-practitioner disconnect seems to only get wider. However, I am optimistic about the growing interest in data-driven HR (i.e. People Analytics). As more companies value applying outside research (considerable though too often ignored) and conducting internal research, the gap should narrow.