Next step of the "what do you mean, maximum minecart speed?" programme has been a smashing success. After some preliminary testing putting ballast into the pushing carts to increase their weight and thus momentum, i found that this totally works. Very good!
Time to take it to the next level: two-tier acceleration.
Four accelerator coils, three collisions, one ultra-fast minecart - over 8,5 tiles per step, i.e. more than three times the claimed maximum speed.
Preparation: four minecarts get accelerated, each by its own impulse ramp cycle. I don't know the exact speed at which the carts stabilise, the results say it's over 250k. The minecarts used are
one bloodthorn, containing a bloodthorn bucket - 53 kg
two palm - each 27 kg
one willow - 15 kg
All carts are dropped into equivalent spots in the circuits by hatch covers linked to a single lever, guaranteeing absolutely synchronous circulation.
Once all carts are circling nicely, a second lever gets pulled, opening all four doors at once. Again, synchronous operation is absolutely required, otherwise, you'll get garbage.
First-tier collisions:
on the southern track, the bloodthorn (older) and one palm minecart (newer) collide. Due to the layout of the track, the newer cart will receive the full momentum of the older cart. 53kg*~250k/27kg gives a resultant speed close to 500k. The bloodthorn bucket will be shot out of the minecart and must be collected and returned to the cart later.
on the northern track, the other palm minecart (older) and the willow minecart (newest) collide. The willow minecart moves off, at about 450k speed or so.
The older carts in the respective collisions come to a stop, the newer carts move off to the east and enter the second-collision track, the north-south track.
Second tier: The moving palm (newish) and willow (newest) minecart collide on the north-south stretch. Once again, the older cart transfers its full momentum to the newer cart. The resultant speed, assuming 500k on the palm cart, would be 500k*27kg/15kg, about 900k.
In fact, running the process does get this result - the willow minecart moves twenty-seven tiles in the first three steps after the collision.
To track the progress of the cart, it then is set to cycle through the closed forty-tile loop off to the northeast. Interestingly, it appears to be subject to massively increased friction: it moves 27 tiles in the first three steps, exactly eighty in the ten steps after that (avg. eight), takes 26 steps for the next 160 tiles (avg. just over six) and 22 steps for the next 79 tiles (avg. below four). After that, it appears to have braked down to terminal velocity and decelerates normally. On a first very rough guess, it appears that a supercharged cart loses something like 10 000 velocity per step when moving over floor. Ballistically launched carts appear not to be subject to superbraking (a cart launched after an ordinary binary collision reached a height of 47z when it crashed into the map border, would have gone higher for sure), so an optimally calibrated second-tier cart could theoretically be able to reach an apex of about 300 z and jump a distance of well over 1500 tiles horizontally before falling back down to the zlevel it launched from. 768 tiles are the full length of an embark area (16 embark tiles).
Next stop, third-tier collisions. I wonder if i'll get fourth-tiers to work.