The assumption implicit in using this as an example of how evolution could not possibly be correct is old and tired. It is basically the following:
"I will find something too complex for you to explain, TODAY, and propose it as an argument that you can not refute, TODAY."
And in that sense it is as good, or as bad, as all before it.
But lets take a seriously look at the weakness of this approach. The most obvious failing is that it assumes, as you did, that the wasp knowingly performs delicate brain surgery which it would seem is ridiculous.
Now I have this wonderful house cat who thinks he is a leopard. If he is in a playful mood he will lunge at the back of my leg and use his front-paw to try to knock me over. Sometimes he will even try to sever my achilles tendon with a playful bite. Does he know about how leopards bring down antelope? No. Does he know how the achilles tendon works? No. Did he go to attack-cat training classes? No. Is his behaviour mysterious or hard to understand? No again. He is playing a built-in attack program that worked for his ancestors. And those that had it survived and reproduced while those that stung the cockroach elsewhere did not fare as well.
One would have to know the behaviour of this particular wasp-cockroach food chain over the past 100,000,000 years to understand the snapshot you observe today. But I can assure you that those wasps that don't get it right don't pass their genetic program into the next generation of wasps.
