Man, whoever told you about that must be well read ;-)

In all seriousness though, I would expect it to work for him, since the placebo effect seems to work for caffeine.

The "anti placebo effect" is the result of a simple classical conditioning. It looks like the strength is just some simple function of the relative timing of the CS and US. It 's an attempt to maintain homeostasis so you don't die.

The "regular placebo effect" is a different beast, and seems to have more or less full access to full cognitive capacity. The information about the contents of the pill gets fed 'down' to the "prior" input to your bayesian estimators. To the extent that the priors are strong and the data is weak, the final estimate sent back up can look a lot like the priors. This actually explains a ton of cool stuff, but that's for another comment/post.

The vast majority of examples of this "anti placebo effect" that I have read about involve injections (there was one story of oral intake). I'm really speculating here, but I think that short time frame things like injections would call up the "anti placebo" more strongly than "regular placebo". The first is that the other drug tolerance mechanisms can usually handle things when the onset is slow, but if the onset occurs in seconds, there has to be a quicker cognitive method to handle it. The second is that the signal to noise ratio is much higher with fast rise times which reduces the effect of the regular placebo effect, and increases the ease of learning the anti placebo effect.

Here are a couple sources on the classical conditioning effect on heroin tolerance.