The trimmer's tempco affects the resistance of the entire potentiometer track. Let's assume you've got a really shitty trimmer that has +10% tempco at some delta-T and you're using it in the potential divider configuration. The 'top' resistor in the potential divider (vith a value at some percentage of the trimmer's total resistance) will have a tempco of +10% and the 'bottom' resistor (the other part of the trimmer's track resistance) will also have a tempco of +10%. Each 'resistor' therefore changes by +10% in the same direction for a given change in temperature, and, if used in the potential divider configuration, the changes cancel out giving you the same voltage at the trimmer's wiper. In other words, the top resistor increases proportionally to the bottom one, so the net wiper voltage remains constant.
However, the input bias current of the reference pin will cause the potential divider to be loaded further with increases in resistance, and so dropping (or increasing) the voltage. So you need to use a trimmer value orders of magnitude lower than that needed to bias the chip to minimise this effect.
Also, we're assuming the temperature coefficient is linear and evenly distributed across the potentiometer's track (which if it's a carbon type it almost certainly won't be). That might be enough to be a concern for ultra-precision applications. Cermet and wirewound trimmers might be better.