Isn't actual real-world usage (market share) more important? If 99% of your users already has the resource cached, no request will be fired whatsoever.
That's not the end of the world as long as the CDN DNS + server connection overhead is reliably better than your own but if you already use a decent CDN that's not a given and it might prove a de-optimization if the resource in question can be delivered in less time than it takes to perform an otherwise unnecessary extra DNS lookup and server connection. For high latency networks that's worth monitoring and reviewing closely as it's often a net-loss.
You probably want to devise a scoring function based on likelihood that the resource is loaded already and cost if it is not and rank your CDN options based on that.