In order to allow us give you a more precise answer, you need provide more details concerning behavior of signal to be acquired, such as its variation range along time and also the rate displacement.
Even a long term counting could improve precision. For instance, if you consider overflowing n-times a counter on its entire magnitude ( e.g a 8 bits one counting from 0 up to 255 ), and divide it for elapsed time, you can get as result an extended precision number, which could compensate mentioned loss on pre-scaler.
+++