I've done some intensive search for published benchmarks and found resources on the net where was claimed to have benchmarked several J2EE compliant AS against each other. But it is as always there were apples and oranges compared (or pears as you would say in german).
Benchmarking is rocket science and a great deal depending on your specific application and AS tuning.
If you want a meaningful result, do it by your own, with your own application.
From a lukewarm point of view both have their assets and drawbacks and in general I would not make my decision on raw performance. I.e. to bring another buzz word in play you may take TCO (total cost of ownership) into your consideration: You can compensate lower speed with more instances which may be less expensive if you don't have to pay licenses -- or more expensive if the hardware, operation and support cost is higher than the cost for licences.