Your calculation is right. (Other than rounding off of 1024 to 1000, the LSB is 3.64 degrees/second
However, the noice entering the system, vibrations from the motors, noice from the gyro itself (the internal 30 kHz osc) and ADC noice causes dithering.
This improves sensibility alot, and as you can see on the video it works
I later used a red LED instead of the 2.5V reference, giving 1.8 V and 2.62 degrees/second. This did not improve flight at all. However, it is incorporated into the second version of the controller, coming soon. It will also have pots for gain adjusting, instead of fiddling with the software.
I can, by biasing the ADC with 0.5V, AC coupling it to the gyro and use the internal 1.1V reference, get a sensibility of 1.6 deg/sec. Or even use an amplifier, but then the low part count starts to rise!