I modified his test script. And posted it in that same post where I criticized the point. When I didn't modify it my times were much much worse on sql server 2014.

It's maybe 10 posts up from here. I didn't change his function I just changed where it was being called to add the traceflag. This increases the randomness of the result as was noted.

I wouldn't really call it a flaw in sql server, you're not supposed to be able to call non-deterministic functions from a inline-table value function, but we currently exploit a trick by using a view. Sql server optimizes the result and considers it a runtime constant. 2008R2 and below don't do this.