Mindboards

I was just using commercial products to point out that you can't expect a $30 toy to be perfectly accurate.

The HT gyro is completely analog. The bias is susceptible to power supply voltage changes (like when you run a motor). Being analog, it is limited to the 10-bit ADC of the AtMega48 co-processor in the NXT, and also to the update rate of 333Hz (not sure that an NXT integration program would be able to keep up with that anyhow). Each change in the raw ADC value represents 1 DPS of rotation (low precision). All of my HT Gyros have a different bias.

The dIMU gyro is digital. It has 1/16th, 1/64th, or 1/128th DPS precision (depending on the range mode). See my report on the dIMU that I wrote a couple years ago, around the time of the dIMU release. As far as I can remember bias is extremely minimal, so calibration/offset hasn't been necessary, at least for my applications (primarily segway type robots).

I don't have any mindsensors gyros.

If you want to try to get a more accurate integration with the EV3 Gyro, you could read the rate value and integrate it in your EV3 program.

that integration thing is what I meant , that was exactly why I asked for lvoc's software program which has been used for comparison.

There is no integration going on in the code I wrote to test the gyros. The ouputs are straight out of the sensors. I just uploaded the code at:
http://www.robotnav.com/gyroscopes/
It was compiled using the Java and the LMSASM tool found in the source code

thank you Lauro, that is good to know.
Then there might be a little chance to improve the accuracy of the Lego gyro angle values by integrating the rotation speed by one self (regarding to my experiences with the HT gyro continuous loops by 100 Hz are always fine).