However in this case what can the users do that can be different to their plan? The most basic example would be logging in, there are not that many execution paths I can think of for that one.
As for the integration / reporting during gameplay, this is all done via a strictly defined API which is called by the game client in relatively predictable ways (game started, player does X, authentication ping every x seconds, achievement unlocked), unlike a web-app where users can do whatever they want the API usage flow is basically controlled by Blizzard.
That is why I am not impressed with blizzard on this one; they control basically the entire use of this API apart from one thing, the number of users trying to use it at any one time, which is there the load testing plan should have worked.
I'd think there are some discussion-worthy points here. For instance, I'm not proud of it but at a former employer I talked them out of load testing an app, which then had serious scaling issues. I learned my lesson the hard way.
If your load-testing does not prepare you for the worst then your load-testing plan is garbage.
They already knew how many copies had been pre-ordered and could make a pretty good guess how many copies would be sold and activated on the first night. Take that worst case estimate, now double it and test for that.
"But the cost of supporting worst-case scenarios!!" some may cry. This is where rented servers / cloud setups are useful for elastic scale without breaking the bank.
There are companies that can simulate load from users across the globe, I have no doubt that blizzard would have the connections / influence /cash to set-up a kick-arse load-test system.