Oops, my bad on the math, it should be 6 ranks each of Bluff, Listen, Search, Spot.
So +3 deflection for total AC of 24. Not shabby. Despite the lower HD, I have a hard time seeing these as much less dangerous than a regular wisp. Possibly a strong CR 4?
Now we just need some great flavor and tactics. Let's add to the tactics, right at the end, "Desperate Will-o'-Deeps will use their sparks to ignite trapped pockets of gas or trigger landslides, counting on their speed and agility to keep them from harm." That sound ok?