If whatever Bandler does is producing verifiable results, shouldn't it be at least an explicit goal of science to find out why it works for him, as opposed to whether it works if you throw an NLP manual at an undergrad?

YES!

Personally, I wouldn't take Bandler very seriously because of the whole "narcissistic liar" thing and the fact that the one intervention of his I saw was thoroughly lacking in displayed skill (and noteworthy result), but yes, you should look at the experts, not at the undergrads handed a manual designed by the researcher who isn't an expert himself. It's much better to study "effectiveness of this expert", not "effectiveness of this technique". I'd just rather see someone like Steve Andreas studied.

I know from personal experience that even people with good intentions will strawman the shit out of you if you talk about this kind of thing because there's so much behind it that they just aren't gonna get. Ironically enough, Milton Erickson, the guy who Bandler modeled NLP after, allegedly had this exact complaint about NLP ("Bandler and Grinder think they have me in a nut shell, but all they have is a nutshell." )