Answer
Bucket counts add up across servers, regions and time, so any percentile can be computed afterwards for the whole fleet, one endpoint or any window. A percentile computed on each server cannot be combined. The cost is precision: an answer is only as exact as the bucket edges around it.
Why
Put bucket edges near the SLO threshold (an edge at exactly 300 ms makes "share under 300 ms" exact), or use exponential histograms, which keep the relative error small at every scale.