MX-64 asymmetric torque problem in current control (p2.0) mode & goal torque control malfunction (p1.0)

I am using eight MX-64 units (ID: 0–7) connected to an OpenCR board.
The OpenCR board has two RS485 terminals, with four motors connected in series per terminal.

To test torque control, I used Protocol 2.0 and applied a Goal Current of +20 and -20 to check if the motors would rotate at equal strength in opposite directions.
With +20, the motors appeared to rotate at a normal speed, but with -20, they rotated in the reverse direction at about one-tenth the strength.
According to the Dynamixel SDK application’s data, the measured current was similar in both the +20 and -20 cases.
This suggests that the current command is being properly transmitted.

The issue is not limited to a single unit — all eight motors show the same symptom.
I would like to understand why the torque becomes extremely weak when a negative current is applied.

Due to this problem, I switched all the motor firmware to Protocol 1.0 and controlled them using the same Goal Torque command.
In this case, the motors rotated with similar strength in both directions.
However, one particular motor spun 2 times stronger and faster than the others.
The current reading for this motor was also much higher compared to the others.
The motors are only mounted with horns, with no additional load, and are rotating independently.
I am curious to know why this is happening.
I have already confirmed that all eight motors have identical values in their EEPROM regions.