Here's mine currently. Sorry if it sounds like a bitch. It's more of where I am with the program.
1/ 64 bit has to come out soon.
2/ Multi object populations with more random variance, like tilt in Y, and colour shift with fractals.
3/ A better lighting model for clouds. Even a setting for depth of shadow cast would be useful, rather than just on or off, or via how dense the shape looks. There are some magnificent layers of clouds outside my window today, with some fantastic pearl like qualities, which I know I can't model, because I know my layers will create too much shadow, and the clouds themselves will not project light.
4/ More control on cloud shape. The cloud node takes over, and has a habit of 'crunching' cloud shapes too much. I'd like to spend more time with this, but constantly get thwarted in trying to creat scale within the cloud depth restrictions, ie I'd like to set the percentage of flattening of the base of the structure, and I'd also like to control the coverage through the height of a cloud. I had an idea of just being able to modulate depth and altitude of the cloud layer, but as it is, the calculation requires a set boundry to work in. That's ok, but we really do need to be able to put more variation in easily.
5/ Caching. There has to be some sort of intelligent caching of GI AND populations. Matt has already aluded to that populations can be switched to real x,y,z co-ordinates rather than population x,y,z. This would allow for travelling population areas to be locked to the camera, and that only the newest parts of the landscape revealed to the camera are calculated, rather than having one massive area, with it's calculation baggage, per frame. I'd happily let a machine work the hours to work this out to file. It would save on render times, memory constraints, and edit time.
I don't really care about the animation module, because camera keyframes are irrelevant in this package. I can set up a landscape proxy in Max or Maya, and then ship over the camera data to render out my plate. Which is what I do at the moment. More control within the nodes via an A,B,C matrix is actually way more exciting than having a graph/dope sheet editor.
Anyway, my tupence thrown into the well.
Cheers
Jon