Different 4D Captcha model experiments

One common complaint with the first batch of 4D captchas came from people who couldn't touch type. They would look down from the screen to find keys and loose their place in the rotating animation. When they looked back up to type the next character, they'd have to wait for the animation to rotate around again.

An idea I'm trying to flush out is to keep each letter rotating but in a fixed order.

It's probably easier to just show the model rather than try to describe it.

Each sphere would rotate in place.

Here's how the basic mode looks without too much variation. BTW, I use POV-Ray + python to do all these renderings.

Rendering of this model using 100% transparent spheres.

A. Basic - no sphere hint

You might notice a few things different about this captcha from the first batch (besides the obvious spheres model).

1. Camera "bounce" isn't as jarring when the camera reverses direction.
2. Image contrast has been increased. 

Again. 4D Captchas are hard. That's the point. They should engage and work your Visual cortex. The question is, if your really look at them for a couple of seconds, can you read them reliably?

Experimenting a bit more with the scene. Here's a rendering with some turbulence on the hex grid.

B. Basic - turbulence hex grid

The thought is that the straight lines in the Model A might be too much of a clue to cracking algorithms since all the characters are curved.

Rendered with the spheres 75% transparent to give a hint of the underlying shapes. Does it help newer users understand what they are looking at or just get in the way?

C. Basic - Sphere hints

These are a bit more experimental. I think a smart ORC could reverse this one, but it is easier on the eyes.

D. Non hex model

This is a mix of the water and the background.

E. Combo

I'm not sure which is a better approach. I like D, but I think it would be trivial to crack with OCR.

Any opinions are readability beyond the typical "these are too hard" reaction?