First of all I must say I don't believe in these kind of unpredictable systems: rarely doing this 'select randomly the process from a set of processes' works better than using the best process in the set.
But I don't think this applies:
But terrorist organizations -- especially those employing
suicide bombers -- have very different goals and incentives
from those of smugglers, fare beaters and tax cheats.
Groups like Al Qaeda aim to cause widespread disruption and
terror by whatever means they can, even at great cost to
individual members. In particular, they are willing and
able to sacrifice -- martyr -- the very lives of their
solders in the service of that goal. The fate of any
individual terrorist is irrelevant as long as the loss
contributes to terror and disruption.
Training a terrorist has a cost, and he should succeed the "fate of any individual terrorist is
not irrelevant". The terrorist group does not have an infinite number of terrorists (as he correctly concedes in the next paragraph).
So random screening works, not because that influences the behavior even of those who aren't checked, but because makes executing the attack more expensive to overcome the possibility of being detained in the random test.
Of course random screening is not as good as full screening, but from a realistic point of view is the only thing you can apply without shutting down world economy.