I think the refinement depends on the objective. For example, for flame speed calculations, a few hundreds of grids may be sufficient. On the other hand, when we wanted to compute local temperature overshoot, we needed thousands of grid points in one of our recent studies. See the supplementary Fig. S1 in Combustion and Flame 263 (2024) 113397.
I am also interested to get more insights on this topic.