I've been having trouble with this 64-Bit-Issue also. It was not mentioned in the Early-Preview-Readme, but then was added to the final readme. According to my mail correspondence with AMD, not even all of their own guys were aware of the 64-Bit-issue.
After the initial report on the AVC (Avivo Video Converter), I've tried some encodings under XP (32 Bit of course this time) and was getting strange results again.
Under XP, the GPU was used (albeit not very much), got warmer and the transcoding speed scaled with core clock on a given card (4670, 4850). But surprisingly the same settings proved to be faster on a non-supported card like 3870 or even X850 - with the app obviously running off the CPU (which was dualcore, mind you!) only. At first, I thought this was due to my ub0r-anti-leet mobo with the PCIe connected only via 4 lanes from the southbridge (asrock 4core dual sata2) and thus more or less my fault but then the same proved to be repeatable on a more modern setup with an X48-Board. Plus, the ALU-count didn't seems to matter as much as their freq, because a 4670 with fixed clock at 750 was faster than the 4850 at a fixed 625-MHz-tick.
Maybe the AVC uses only one SIMD (or no SIMD at all but other portions of the GPU?) at the moment - dunno. This point seems to be true for Badaboom also, since it fails to scale according to even GFLOPS, let alone the architectural differences between G96/G92 and GT200. I think right now AVC is more a proof-of-concept than even a techdemo (which i would consider Badaboom), sadly.