Sadly JSONP continues to flourish and so requires substantial work to get some semblance of good performance for large datasets. JSC explicitly checks all program code for the common JSONP idioms and pushes them down a completely different execution path (that is equivalent to simply calling JSON.parse) :-/

Now run two fetches at the same time, and watch it get the data mysteriously mixed up 0.1% of the time.

simple and clear! what's next?