Yes you can accelerate it, either by optimising the ARM code used by mplayer (and I understand that there is quite a bit of room for optimisation there) or by passing off some or all of the decoding to the DSP (by writing a DSP task and ARM-side wrapper).

It would be interesting to learn why the built-in media player is apparently slow - it uses GStreamer to wrap the DSP decoding. The question is then whether to write an improved GStreamer + DSP task or to use the DSP without GStreamer (e.g. if that's what's causing the slowdown).