Thanks, you have been extra clear!
I was a bit off mark about the torque, at 2.5A I was expecting 300g/20mm instead of the more realistic 250g/20mm, but it is likely still more than I would need.
Let’s put it this way: the robot weights 500g and its weight will be equally distributed on the wheels and on a third point of contact with the ground.
Let’s say that each wheel is supporting half the weight (which is impossible: the third point of contact is needed): each wheel would have 250g of load.
At that torque in theory each wheel (with a radius of 20mm) could produce 250g of force, but it could not transfer that force to the ground because the friction coefficient will be less than one: the wheel will be overpowered and it will start slipping.
More briefly: I will need to throttle the torque anyway, to avoid wheel slipping!
At least on paper, but the numbers look solid and past experience with similar robots (with plain DC motors) confirms it.
I am quite familiar with the “stuttering” caused by thermal and overcurrent driver protections kicking in when a robot stalls.
Several of the competitions I participate in are for autonomous robots built entirely out of Lego pieces (with Mindstorm sets), and it is really easy to trigger those protections.
Regarding speed, my real target is just “a bit more than 2m/s”.
The main competition I am working on with this design is this folkrace.
The currently winning robots run the straight parts of the track at about that speed (2 m/s).
The hard part of the competition is the software, particularly autonomous driving with other robots racing around you, and detecting whether the robot turned around after an accident.
Another competition where I’d like to use this controller/motor combination is this mini-sumo.
There the arena is 70cm, there’s no way to achieve stellar speeds in such a small space.
Torque would be more important, I could decide to reduce wheel size to 36mm (just a tiny bit more of the motor diameter!) but I’d still like to keep the direct drive configuration for simplicity and efficiency.
Or I could add a third motor (and wheel), with an “in line” positioning.
I will do tests and decide.
BTW, I am part of this team, we are really used to these competitions, but this would be my (and our) first brushless design.
Finally, about heatsinks, I know I am risking stalls and I take it seriously.
I am considering not soldering the position sensor on the mosquito, and instead use this external board.
I would put the mosquitos on the upper surface of the robot, and I would add the heatsinks and a small fan for active cooling.
In the folkrace bot I also need active cooling (the same small fan) for the main CPU board (a Raspberry PI), so I am used to this.
A small fan weights a few grams but can solve a lot of problems 
Wow, this was long… I hope it gives ideas to other builders as well!