Maybe they do, maybe they don't. My point is instead one of ecological validity (a form of external validity) referring to the extent that experimental findings apply to real world situations. These researchers investigated how anxiety and uncertainty can make us become more radical and committed to religious causes. They did most of this with computers using implicit measures and explicit measures for testing attitudes towards religion. Self-report was used to test whether participants would 'die for their cause' among other things.
Here's the problem: we do not know how or if people who score 5/5 on wanting to die for a religious cause (the participants were undergraduates) are actually willing to do go and risk their life if the religious cause called him/her up and said alright it's go time. Saying you'd sacrifice your life for some fictional religious cause on a computer and actually putting your real life that actually means something to you at risk is completely different from some temporary virtual life you are willing to send off to religious war or what have you.
Think about how careless people can be when playing video games - why would you care if you die, you respawn anyway. It's the exact same problem for experiments that try to test social phenomena: we behave differently when the risks are real vs fictitious. But doing real world research on terrorists or any other form of criminal is complicated if not impossible. For example, there are obvious ethical considerations if I wanted to test how likely someone who scores highly on an explicit measure of prejudice is to commit a hate crime. You just cannot do that (at least where I'm from).
Nevertheless, the problem of ecological validity is an important one and should be pointed out whenever possible, though justified as to why it matters in the context of a particular experiment and the real world situation it relates to.