Here's where the conclusion is unclear : it says "The cost ... was a mere $400,000." However, according to the article's math, that's the cost -per factory-. The policy for the entire company would cost $800,000.
However, that's not quite right either. In order to return $10M total, you'd only have to buy about $334,000 of insurance on each factory (I'm rounding the numbers for convenience). Since the example operates under the assumption that you can always use the proceeds from one policy to pay for the other -- ie, there's never a perfectly simultaneous event at both locations -- I'll take that as a given.
Let's say that Factory A gets hit first (California or Japan, it doesn't matter). We collect our insurance payment of $1,670,000 ($334,000 5) and spend the total amount on additional insurance for Factory B. When Factory B gets hit, we receive $8,350,000 ($1,670,000 5) from that supplementary purchase. So where does the missing $1,650,000 come from? Remember that we couldn't know which factory would be hit first, so we initially bought $334,000 policies on -both- factories ... and that policy is still in effect on Factory B. It returns $1,670,000, which brings our total to just over the $10M target (remember, I'm rounding for clarity). So the total that we need to spend on insurance is ($334,000 * 2 =) $668,000 per year.