HI Emilio,
I just compared the kernel from SDK6 and SDK5, and it does indeed look like there are no changes that could account for this.
I would definitely prefer to be testing with tvout, however I have not yet gotten my hands on the proper daughter card for for my EVM, nor do I have a tv with composite input at the moment. I definitely agree with you that now it seems the major culprit is the clock framework that feeds the tvout capping the max fclk. Like I said, my major concern now is that based on my tests with DPI output on the EVM, the scaling requirements you have now seem to immediately cause a FIFO underflow in the DSS.
That being said, I will continue to look through the patch trail from 2.6.37 to 3.14 to see if there are patches available that improve the clocking capabilities for tvout.
Regards,
Josh