I've been doing a bit of Raptor tuning to get down the CPU and memory usage on large files. I was always afraid of premature optimisation and knew that things could be improved a lot if I did some profiling and cut down the big problems. The results on a 550,000 triple rdf/xml file:
Raptor 0.9.8 Raptor 0.9.9 (CVS)
That's over 23x faster. The improvement is due to a lot less strlen(), malloc()/free() and using a set for ID checking, rather than a list.