On the measurement of the Casimir torque