Hmmm…According to my friend Duane, this is because the longer you roast the coffee, the more caffeine you burn off.

This requires further investigation.

Here is my Google Search on the topic.

And here is a technical blurb from Roast Magazine:

Roast Level

Beyond selection of the green beans, the roaster is commonly thought to control one more variable in the final caffeine content of the beans: the roast level. Popular lore has always been that the darker the roast level, the lower the caffeine content. This is not really the case, as caffeine changes very little during the roasting process. Caffeine has a very stable crystalline structure with a boiling point above 600 degrees Fahrenheit, far above roasting temperatures, which rarely exceed 470 degrees Fahrenheit. This means there is very little change to the caffeine during the roasting process. The minimal amount of caffeine lost during roasting is attributable to sublimation, which is the transition of a substance directly from its solid state to its gaseous state, as commonly occurs with dry ice. Caffeine undergoes this transition at around 350 degrees Fahrenheit. Since coffee is roasted at temperatures above 350 degrees, a minimal amount of the caffeine is lost this way during the roasting process.

Although minimal caffeine is driven off or destroyed in the roasting process, the bean undergoes major changes during roasting. This can confuse the situation because the caffeine content per weight and per volume changes—not because the caffeine changes, but because the size and the weight of the bean changes. Ironically, because the bean loses weight (mostly water) during roasting, the caffeine content by weight increases, but because the bean increases in size during the roasting, the caffeine content by volume decreases.

It is fortunate that there are no requirements to label caffeine content on packages of roasted beans. So many variables contribute to the caffeine content of a single origin at a defined roast level that it is nearly impossible to predict the content without decaffeinating the bean and measuring the amount extracted. Now take differing cultivars from multiple farms and multiple countries, throw in a little robusta for an espresso blend, and you might need to put on another pot of coffee and call an organic chemist.

Here is another blurb:

Here’s my completely non-scientific anecdotal answer:

When I roast beans, if I inhale the smoke that’s coming off them, it gives me a caffeine buzz. That would indicate to me that at least a little bit of caffeine is being burned off the longer the beans cook. I’m sorry I don’t have any hard data for you, though.

And:

I believe that in an episode of Good Eats Alton Brown said that lighter roasts retain more of their caffeine.
posted by Rhomboid at 3:44 PM on February 6, 2007

Caffeine definitely comes out in the roasting process- I used to date a coffee roaster, and if I hung out with him while he roasted, I’d be jittery for hours afterward. That being said, I don’t know if you’d notice the difference in caffeine between a light and dark roast. A lot of the caffeine comes out with the the steam at the beginning of the cycle, and I’m not sure if that continues in any significant during the carmelisation process. The theory makes sense, but I’m not sure if it has real world applicability.
posted by oneirodynia at 4:46 PM on February 6, 2007

My take on it is. I really like Guatamela Peaberry and I recall that it was a lighter roasted coffee and I felt like it had a bit more caffeine, but in a calmer sort of way. It didn’t make me all jittery and anxious like some darker roasts. I do like the idea of breathing in the steam from the first fermentation process. I think I should get into roasting coffee to investigate further.

Dan