This reminds me of another post Yudkowsky made on a very similar topic.

I would consider most of the people here to be informally operating on at least a milder version of Crocker's rules.

I can think of a scenario where the colour red tends to have the opposite effect... http://en.wikipedia.org/wiki/Redshirt_(character)

There are conflicting issues though. There are studies (that I read years ago, and have no link to) that show that consistency is better... that consistent low-level caffeine drinkers are more alert than their non-caffeine colleagues, but less jittery than high-caffeine people (optimum seemed to be 2-3 cups per day).
Associated with that would be method of consumption: concentrated does (espresso) v. sipping american coffee over an afternoon. Using is in a "targeted" manner might fail you: If you are not particularly used to the effects and suddenly drink coffee for short term memory reasons, you might not get the desired result because you'd be too "hepped-up" (to use the technical term...ha!).
If you ARE used to drinking coffee, and suddenly avoid it for long term learning reasons, you might be either sleepy or hit withdrawal.

Re: "Use caffeine for short-term performance on a focused task (such as an exam)."

Really? What about state-specific memory? If you are intoxicated by caffeine during an exam, don't you need to be taking it during the revision process as well?