What needs to happen on the PC is to extend the current video acceleration API's to handle the entire demuxing, decoding, and presentation process entirely on the video card (both audio and video). Right now you can feed a h.264 bitstream to the GPU and it will return decoded frames. But you're still stuck with Windows code handling the demuxing, a/v synchronization, and presentation. Here's where things typically start to go wrong:
1) Proper 100% accurate A/V synchronization is impossible because you have totally unrelated audio and video clocks. Audio or video frame drops are inevitable given enough time. This needs to be handled on the video card using the same base clock.
2) Proper presentation is also impossible because you're dealing with API's and hardware originally intended for games and applications - they can't natively display YUV interlaced content the way a dedicated STB using overlays can. You're also at the mercy of the Windows task scheduler as to whether your frames will even be presented on the correct vsync interval.
The closest thing to proper PC based playback were the old MyHD cards. Until someone else decides to release a similar product - you're going to keep fighting with the problems discussed in this thread. Of course none of this matters to the majority of the users in this thread. Only videphiles seem to notice or care about these issues.