Ideally, the sensor response time is faster than the system (time constant) for stability in PID control etc. if that is what you are doing.
If the system changes temperature faster than that (sensor), feedback control gets sloppy as you are always lagging, always behind, and need 1 second at least to get accuracy.
You can model the sensor's thermal characteristics (time constant) and attempt to predict (add lag compensation) but it never works well for me. The problem is the system gain (that a model is built upon) might change and make closed-loop control go unstable, if you are modeling it. A classic example is a change in airflow past the sensor, that controls a heater. Higher airflow decreases thermistor response time (due to better heat transfer to it) and usually system gain goes down due to the heater being less effective with higher mass flow.
I might be wrong here but I have made a few oscillators by accident.