Peter, At one point during my testing of this system, I did run into the problem that the 10GBe did not output data. The problem ended up being related to flow control into the 10GBe model. Since you indicate that for a frame size of 4096 the system works, this may be your issue. My fully implemented design uses the TxFull flag to pause loafing data into the 10GBe module.
If you wish to send a copy of the model to me, I would be happy to look at it and see if there is anything obvious that should be changed. -Joe Kujawski