"If you're 15 years into that research and you think there is a chance that you'll reach a breakthrough in 5 years, you're much more inclined to want...to pursue it for 5 more years rather than starting over."
Decision A: 5 years (after 15 already spent) = 75% chance of success
Decision B: 20 years (after 0 already spent) = 10% chance of success
In that scenario, Decision A (sticking with your current research) is more rational as long as your estimate of your chance of success is accurate. We know from the article that this is not true -- their estimates were wrong.
The sunk-cost fallacy would apply for something like:
Decision A: 20 years (after 15 already spent) = 10% chance of success
Decision B: 20 years (after 0 already spent) = 10% chance of success
Since both A and B have the same cost and same likelihood of success, you should not consider the time you've already spent. Most humans would, though.
How is that not sunk-cost fallacy?