I don’t believe slow user refresh rate will prevent you from getting accurate results from a custom estimator. Thermal drift and mechanical vibration of the gyro are, probably, going to contribute most of the error to the position (angle) estimate.
Depending on the electrical, thermal, and mechanical noise levels you may need to throw out large percentage of the observations. However, good estimator does more than a simple integrator.
Large robot is not going to wildly accelerate or decelerate over a 10ms interval. If you correctly model robot rotation as a function of current angular velocity and acceleration due to the its own motor power and possible interaction with external objects, then it shouldn’t be that hard to filter out any out-of-range gyro measurements.
Accurate knowledge of the measurement latency, is probably, more important than the refresh rate of the sensor.
It would be great if we could know even approximately delays involved in getting analog measurements on V5, similar to how these threads go into details of Cortex hardware timings: