Had some fun with this myself this past weekend. But, just about got it figured out now:
The performance was dire initially, but I added the module option "optimization=1" which bizarrely optimized for CPU according to the code. But, this got me up from 12k/sec to at least 5MB/s.
I also had what can only be described as crazy IRQ issues. All three of the devices that I would use most (ethernet, dxr3 and DVB card) were sharing the same IRQ. This manifested itself as nasty picture breakup when the network was in use (useful!). Hard coding the IRQs for the PCI slots in the bios seemed to sort it out.
Oh, this is 2.4.21-ac4 on an A7N266-VM.