It rather looks as if many Akamai proxies got poisoned by the root server sending out error pages with a "200 OK" response during temporal failures - as the proxies registered a success response, they cache the request and will not refresh until their timeout is reached.

That is probably due to insufficient tests during a rush to Akamai - it may have worked out in the past while the server was standalone, and did not get immediately obvious until there was heavy traffic, but in the current situation, failing to to return error responses on failure creates a huge mess.