If a script kiddie does something directly from his home computer, then the source of attack is clear and you can just subpoena his home address from his ISP and get a cop to educate his parents or whatever. However, if pivot machines are used, then you need to obtain data from every machine in the chain to trace the source of attack, and doing so by a chain of subpoenas is extremely slow, and compared to that hackback provides a practical (though usually not legal) option to do so.
Also, hackback provides a venue for stopping distributed attacks - if you're attacked from a large number of compromised machines, you need data from them (again, either by subpoena and physical retrieval or by hackback) to identify the command&control servers and possibly other victims.
Also, is it really the case that "virtually all attacks" would go through pivot machines? Certainly the sophisticated ones would, but there are plenty of script kiddies to bring down the hammer on.