I've got an idea that might work and result in all girls having 3 forms. I have no idea if it'll take more or less work than just rendering everything 8 bajillion times but I'd say it's worth at least thinking about.
So 9 * 3 = 27. That's 27 different forms total. That number then grows exponentially when you talk about the different combinations of different forms in multi-girl scenes and the main story. I've got something that might help limit file size and reduce the workload at the same time. I split it into 2 parts for organization sake.
1st part: Static backgrounds. Basically instead of the background being a full 3D render, make it a still image of a 3D render. Basically just a screenshot. Using this for events might be a good idea during talk segments as you can then do what Corrupted Kingdoms does and have them pop up when they talk in a simple 2D image form instead of a full 3D render. CK even does this for the main story in Act 3.
2nd part: 2D image characters. Have each character that has a New Life form (or all the characters if you want to be consistent) have a series of 2D images that they'd start using at a certain point in the game. Corrupted Kingdoms does this for all scenes and characters, when you enter Act 3, as the game allows you to customize the outfits some of the characters wear. So instead of rendering each and every scene in multiple outfits for each character, each character was instead given a 2D set of images consisting of different expressions and poses they'd do while talking thus eliminating the need for a million complete 3D renders.
The only problems with the 2D images would be coding, and the h-scenes.
Coding might be easy, it might be buggy, (but that's coding in a nutshell): if player chose this form for this character, then this set of images pops up when that character speaks. Then, you'd just have to match the different images (expressions and poses) in that set to their lines and you're done.
The h-scenes in Corrupted Kingdoms act 3 are still 3D renders and usually has the characters say something like "Lemme change real quick" to justify them being in a different outfit than what the player chose, for that scene. I'm assuming you'd want unique scenes for each unique form, so those would unfortunately still have to be rendered in 3D, same goes for any multi-scenes (which would result in a decent bit of work, but combined with the 2D images for talking, it should result in less renders overall)
There it is, the essay I thought up randomly while trying to sleep one night. Sorry for the long read, but I thought it might help to at least give you ideas.