The particular temptation that I saw was that they _aren't_ a substitute for a full graph -- they're only a substitute for the information one gets were trying to put so much information in them that one would need to spend time studying them to read them, and that misses the point.
Personally, I haven't yet written anything that seemed to provide a good use for them, so I'm sort of waiting to have a real opinion on them until I actually find a place to give them a proper try-out in a "real-world" situation.
In any case, I do agree that MetaPost is probably one of the best ways to implement them, and ConTeXt's MetaPost integration should make it quite easy to organize such implementations.
To clarify, my question wasn't a request or anything. I see it in the same position as you, a toy used in exploration of some concepts that nonetheless seem to be based on some valid points. One of them being that the data and the text shouldn't be too disparate, floated far afield, and another that the content to presentation ratio should be maximal. These Sparklines cannot replace a full-blown figure but afford to approach data more in the sense of reading than in the sense of deciphering. I asked here because everything about it screamed ConTeXt (also considering that the favourite command here seems to be \input tufte). Thought this meme must have infected some here. - Tobias