Sorry, but these benchmarks are crap, for several reasons.
First (*), using CPU-based decoding, the file does not play in realtime. No word about that in the article; it almost looks like the author failed to notice or understand it.
The Atom 330 (not N330, which does not exist, by the way) is a dual-core w/ SMT and appears as 4 logical cores. This means 25% CPU load in the graph represent full load on one core. The other cores are not used at all - MPlayer is single-threaded.
Second, hardware accelerated colorspace conversion should be activated when gl2 is used for best results. The suboption "yuv=2" does that.
Third, knowing (*) it doesn't make sense to compare power consumption.
The cloverfield sample is technically illegal in the US (and other countries) and can't be legally redistributed without consent of the studio. Even use of the clip can be illegal as well. It's a great clip to test but falls well outside "safe and legal" sources.