This is an old problem with modifying servos for continuous rotation. After the mod is done and you are actually using the servo, you notice that it doesn't stay still at all times. This may be because you glued the pot a bit off center, but if it wasn't moving when you glued it and the servo functioned properly after the mod when you tested it, that is not the problem.
The real problem is the construction of the pot. It varies the resistance significantly in a small area where the swipper touches the resistive band. So it is sensitive to temperature, voltage, lenght of use and so on. To make sure that is not happening, you would have had to replace the pot with a multiturn pot, find the turn where the motor stops rotating, then keep turning the pot until the motor starts rotating again, then turn the pot back until you get half way between motor stop and motor start to be sure you're on the proper spot. That will make the servo stay still perfectly and it's the best method I know off.
Another method lays in the software. Everybody is using delay_cycles for the servo PWM, which is not perfect. Depending on the servo, you need to be more precise. The only way to be exactly on the spot for that particular servo, is to use a timer to generate the PWM signal. If you set it up so you can use microseconds to set the pulse, you have a lot of room to play with the values in the same fashion you would play with a multiturn pot. Set the pulse width to be in the middle of the area where the motor is stopped and you will be fine (well, because the pot may change it's value because of temperature, etc., you may still need to tweak the value from time to time).
Well, you get the idea. Good luck with your project!
