I have both ways implemented at my plant, one counts pulses, the other integrates and there isn't a noticeable difference between them.
However, we use this for batch loading, so we run it for maybe thirty minutes and then start over, so we don't really see any slow drift in the reading.