No subject

compared to the PCI-E version of a card, I'd expect the PCI version to suffer from performance and, potentially, compatibility 
problems. The bus interface bandwidth is greatly reduced, certain features like non-coherent system memory accesses are unavailable, 
32-bit PCI cards are unable to address memory above 4GB, limiting the pool of memory available to it, etc.. The graphics driver 
stack may also take paths not as well exercised.
the limitations I pointed out are not NVIDIA specific, they are limitations of the PCI bus. In terms of adequacy of the 32-bit PCI 
bus's for streaming video, I don't have a good sense for the amounts of data being streamed (Stephen will have a much better idea), 
though I'd expect them to vary. Available bandwidth and latencies will also vary with the bus topology, the presence of other 
devices on the bus in question, their type and DMA characteristics, etc.. It is my understanding that the GPU(s) in question are 
PCI-E GPUs bridged to PCI, which has additional implications.
Due to all of the above, I can't say whether a PCI card would satisfy your requirements. As to driver issues: if PCI specific bugs 
are identified (I'm not aware of any at this point), I'd expect us to want to resolve them, but chances are they'd be prioritized 
behind more generic ones.

Profiles in place:
<      W: 1920 H: 720, decoder: VDPAU, renderer: VDPAU, Deinterlacer: Advanced 2X
>=    W: 0 H: 720, decoder: VDPAU, renderer: VDPAU, Deinterlacer: Temporal 2X