This one is one of their less commendable efforts: yes, there was a discrepancy between what Heroku's docs said and the actual behavior of Heroku's routing. No, RapGenius's simulation is nowhere near correct: multi-tenant resource allocation is a pretty difficult stochastic optimization problem, and their simulation does not take this into account at all.

Heroku should own up to - and has owned up to, in my opinon- their communications fiasco. But I really, really hope the linked RapGenius article doesn't mislead folks to believe intelligent routing is easy (Tom Lehman, the then CTO of RapGenius, admits that he himself is a novice about distributed locks and such: http://s3.amazonaws.com/rapgenius/1361210233_heroku-adam-con...)

EDIT: minor grammatical clarifications

I did too. Learned a hell of a lot from them that's for sure, lol. Their Engineering team does a lot of, in my opinion, cool things and I really appreciate them taking the time to write about it.