Pipelining a recursive equation

Actually, rereading the suggestions above I didn't appreciate at first that thepancake's comment about sampling also amounts to this necessarily being a decimating filter. So I can build a 3-stage pipeline as long as the input data is downsampled by a factor of 3... (and if I decimate more than that I can even reuse the mulitpliers as suggested and get to a very small very fast design) and perhaps the affect that will have on the filtering characteristics can be compensated for in the coefficient calculation... so perhaps decimation will save me!