Seconds (judging by the source code), and I assume the variance comes from the time it takes the server to respond + javascript delays. It's interesting to see the significant variation and an almost 100 ms difference between what my browser reports and what the chart reports. (The server responses were pretty consistent).