A major failing of sparklines in practice is the lack of a visual benchmark.
When you don't have a fram of reference, it can be hard to tell what directions the lines are moving. You end up with an effect like this illusion: http://www.internetgamesfree.com/games/images/illusion_68.gi... where your perception of the line deviates from reality.
The "normal range" line he shows does a great job to fix this.
We've started adding these selectively to internal tools to supplement more complex tools aimed at depicting progress or trend visualization. It's helped a ton, and goes a long way toward turning indecipherable IT goon geek speak into something normal users find helpful (along the same lines as somethings like Google's App Status Dashboard (http://www.google.com/appsstatus), the Stashboard (http://www.stashboard.org/), or the Salesforce.com status board (http://trust.salesforce.com/trust/status/) do for red/yellow/green statuses at a glance.
I must admit I don't totally get what sparklines are and what they aren't. They seem quite recent, but the examples I see don't seem that unusual or new. Is it a new name for a concept, or is the concept itself new, or is it a particular take on or way of using that concept?
One of the main things that distinguishes sparklines from other small graphs is that they are "word sized" and can be integrated into a line of text. Because of this, the graphs tend to be wider than they are tall and also have a very clean appearance to them (not lot of background color or other "chart junk" as Tufte calls them).
I would say what you linked to would not be considered sparklines from the above.
I use them in one of my iPhone apps (Dayta) and it's an easy way to provide trends at a glance and make a simple design look much better in a few lines of code.
These rules are useful heuristics, but a good graph has a purpose, and its parameters are chosen accordingly. So, while Tufte may favor a wide sunspot graph to emphasize the weak downward slope, my advisor (a solar scientist) would have slammed me for that. We already know that there's a weak downward slope. What we care about is the absolute magnitude of the peaks. Tufte's graph is useless here.
I'm impressed with the quality of the comments. They appear to be not only moderated, but also summarized in one line by Tufte, who then responds in a separate posting. Very pleasant and skimmable format.
When you don't have a fram of reference, it can be hard to tell what directions the lines are moving. You end up with an effect like this illusion: http://www.internetgamesfree.com/games/images/illusion_68.gi... where your perception of the line deviates from reality.
The "normal range" line he shows does a great job to fix this.