Windows seems to have more hysteresis or longer sampling period on their estimates. I can light the CPU up, and battery won't drop from eight hours to two until it's been that way for a while.
I still think something like the battery over time graph is a better solution than a single number with so many assumptions built in that are opaque to the end user. The graph can express "you've not been using a lot of power but whatever you started recently has caused it to drop a lot faster."