Inspection of onLoad and tffb appears to show a few anomalies w.r.t 20-80 rules.
Enclosing three random cases, where the reason appears to be redirects? In the cases, the ttfb >> onLoad. All the cases have 1-2 redirects before the actual HTML is served.
pageid, time, status, type, format
pageid IN (87277828, 87291432, 87267343)
ORDER BY pageid
The ttfb appears to include the time taken by all the HTML requests (including redirects), but the onLoad appears to be for the last HTML.
In such a case calculating ttfb/onLoad for the backend/total time ratio may include a bit of error. Using the ttfb of the HTML with 200 status (by looking at the requests summary table) may be more accurate.