Perhaps an interesting analogy, but offers nothing in terms of how to improve estimation. The only thing I can take away from this is to always add contingency to an estimate.

Instead of four hours, this could have taken half an hour by properly exploring the problem and mapping out a solution first. Everyone seems to shit on drawing up an architecture first and wants to immediately start coding, but taking a little time up front to think through the problem pays huge dividends in the long run.