I mentionet some problems in the project descriptions:

• The backlash of the part is still there - the culprit my be the triangle shape of the teeth. A more trapezoidal shape with steeper angles might reduce it further.
• The stiff rotor has to be printed to the correct size - a design with the possibility to correct its length could prevent reprints and shorten the time used to fit it all perfect. 
• A stiff rotor also has problems with deviations from the right diameter - some kind of flex could prevent the motor from stalling if there is a little bump in its way.
• If you want to change the gearing ratio you have to reprint the biggest parts of the design -  the cup and the housing. Some sort of exchangeability in the housing would make iterating much faster and cheaper.

Teeth design:

The german company Harmonic Drive, a leader in design and manufacturing of strain wave gears, promote their "S-Tooth" design as an unique solution. I suspect that there is an active patent to this. A Sinusoidal shape might be infringing that patent, I have to check that on www.register.dpma.de  (website with search function of the german department for trademarks and patents)

As mentioned, a trapezoidal shape should be sufficient for minimizing backlash. 
A inside edge angle of, or less than, 20 degree are the next designs I am going to try.

Flexible Rotor:

I got some ideas for this point, as it can be seen in the following Picture:

The stiff length compensation solves the problem with reprinting the rotor parts for a perfect fit, but cant compensate for artifacts that cause a slight change in diameter (STL faceting)

That's where the flexible length compensation can shine. But a  slight difference in spring rate from top to bottom part of the spring will cause the ball bearing not to be parallel to the flex spline what is going to cause unnecessary wear and tear.

Inspired by spring loaded peristaltic pumps, the third concept is able to compensate length for a good fit and differences in diameter by facets, but is also stiff in the plane for the ball bearing to be parallel to the flex spline. The Drawback is that it needs more standard parts that have to be bought (spring and bolt for the pivot)

Ultimately concept four fulfills all requirements and doesn't need more standard parts than the current design. The right dimensions for the hinge and the spring have to be found though.

Exchangeable gearing ratio

The goal is to minimize the wasted amount of Filament and thereby accelerate design iterations if another gearing ratio is needed. As the flex spline has to be, as the name already implies, flexible, I don't see a way to only exchange the teethed area of the part. But the biggest part, the circular Spline and Housing, has potential to allow a separation of the to functions into two parts.

As a big thread could unthread itself, depending on the direction of rotation. 

A dovetail like structure, circular patterned around the center of the gearing, would allow a change without the downside as mentioned above.