I'm hesitant to embrace the smoothed floor idea without some clarifications. If water collects on smooth rooftops, it will eventually flow off the side of the house, which is fine, but then the water will radiate outward from the house in a circle. This would look silly. The ramp condition isn't enough, as just putting a little ramped steeple thingy on top of the house would lead to the same problem. If the frequency of water placement for smooth surfaces is decreased vs. pools, you don't solve the problem until the frequency is low enough that it's essentially an aesthetic feature, which doesn't solve the water collection problem. Making water soak into the ground would solve the rooftop issue, but then player-controlled water that you want to cover the surface would have troubles -- the core problem is that rain water should be less than 1/7 units, but the game doesn't operate at a finer resolution, and rain is only realized as a flow when it is around the scale of regular water (ie a river).
There are all sorts of different angles that could be used to attack this problem. Raindrops don't place water all the time, so depression detection could be pretty easy -- once the raindrop says that it is one of the 0.1% of raindrops that actually tries to place water, it could just flood out a very small distance searching for ramps and walls, and if it gets a certain percentage of the frontier blocked off, it can place the water. Since 0.1% is very small, and the flood out just needs to check 25 squares or so, there would be no lag from this -- it could also be precalculated and stored as a flag, updated with mining etc., if this were to be used elsewhere to the point that lag becomes a problem. This would be enough to get aqueducts and player-made pools to fill up -- adding the smooth floor requirement would spare some trouble from existing map features, but that would just be something to test out.
On the other hand, there's a certain attraction to having water fall off of rooftops and so on. As I said above, this is tricky with the current resolution of water. The "wet bit" idea just puts off puddle formation by one step, so it doesn't address the main problem of circles flowing out from rooftops, but changing the idea a bit might work. The goal might be to somehow let the wet areas spread out until they find a depression square as defined above (even if its underground), at which point they can consider adding one unit of water. The issue here is how to judge the wet frontier -- tracing a raindrop from the roof down through wet areas to a depression underground would be expensive, even if it only does it for a few drops. Combined with aesthetic puddle formation from a very low placement frequency on any smooth stone, it would work, I think, but I don't have an angle on the wet frontier at this point. It would be cool to be walking up out of a smoothed stone tunnel and have the entrance inside be "wet" though, since some rain ran down there, or during a heavy ran have the lowest parts of the tunnel fill up with some water. I haven't thought much about the ground saturation and so on, and combined with aquifers and all that, there might be more to do here. CPU issues abound when specifying the aquifer amounts though.