You reminded me of updateing the infos at the end of the script. The way you tried works with nvidia and intel (there you need -2 script) but not with fglrx. fglrx only works with -vo vaapi:gl[:reflect]. Don't know when this will change.
2nd hint: if it still does not play check mediainfo if it is h264 L5.1.
I'll have to find some more taxing clips, but in my experience so far, this has pretty impressive results. I've been trying some of the various h.264 clips (trying to find the highest frame size and bitrate ones I can) on http://www.h264info.com/clips.html
This isn't a comprehensive, or particularly scientific benchmark, but it gives an idea of the difference in processor (CPU) usage between Xv and XvBA. I'm using htop to monitor mplayer, btw.
Phenom II 810 (2.6GHz), MSI 790GX-G65 using onboard HD3300:
I am Legend 1080p (really 816p):
without XvBA: max 77% CPU
with XvBA: max 3% CPU
Coraline 1080p (from the link at the top of the page at Digital Digest)
without XvBA: spiked at 99% usage 102secs into the clip, mplayer complains system is too slow, but for the most part was 30-70%
with XvBA: %2-3
So far any of the others I've played back give the same results: no more than %3 usage. If anyone knows of a particularly strenuous clip I'd like to try it.