If you want to use computers to save anime, I don't think 3D is the way to do it. Although not to the same as extent as occidental animation, like all 2D animation, anime is significantly non-Euclidean: Features that look the same from multiple angles, features that move around objects purely depending on angle, proportions that shift regardless of position, mismatched scales, much of the exaggerated squash and stretch, etc.
If you're at all interested in the technical side of anime I HIGHLY recommend watching the following panel presentation on its technique and evolution. It's a bit over an hour and a half long, but very good:
youtube.com/playlist?list=PLuPNSyztkHPqV-M4ePSmN_2BLhONu0WLg
Design style aside, using 3D often means creating a lot of detail that isn't essential to a given scene because it's obscured or distant. And then animating it demands meta-information like materials, inverse-kinematic "bones", soft/rigid body distinctions, and similar.
To be sure, these disadvantages become advantageous depending on your stylistic goals: If you want highly dimensional, solid geometry and animation, and reusable or similar looking high-detail elements, it's obviously great. But even the state of the art in "using 3D to fake 2D" from Disney/Pixar leans WAY more heavily toward the 3D side by technical necessity:
youtube.com/watch?v=TZJLtujW6FY
The correct approach would be to focus on the most expensive and least artistically demanding part of 2D animation, inbetweening, and leave as much of the rest of the process intact as possible.
As you can see from the following list, automatic freehand tweening is so neglected as a research subject, the primary feature of today's digital "animation" software for freestyle is simply to make inking and painting each frame easier:
2danimationsoftwareguide.com/
Currently, there are two basic types of software that I suspect could be drawn from for a solution. One is Flash and its ilk, which rely on 3D-like bézier curves, which means either extremely labor-intensive modification nearly every keyframe for freestyle animation, or falling back on "paper doll"-style "limited animation" using separate objects for fully automatic tweens. The other are Elastic Reality's descendents, which can automatically generate hints for "morphing", that are then easily edited, but due to the arbitrary nature of the bitmaps it operates on, it has little information to guess similarities between keyframes. By combining vector-based drawing and some kind of pattern detection, with the primary focus being the minimization of human input for hinting keyframes, I think the full capability of 2D animation could be brought down to a similar budget and production speed compared to modern 3D animation.