I'm building a generic text processing engines which creates parser from XML definitions and I'm facing a kinda big problem : reg exp entities are just SOOOOOOOOOOO SLOW...

You see, I knew that reg exp ate a lot of CPU time but I barely noticed a difference until I loaded a 1.2Mb file (15k lines). The cost of reg exp is anything but affordable :
  • RX enabled : processing lasts more than one minute
  • RX disabled : processing takes only 3 seconds
Is there anyone knowing a smart trick to make reg exp decently fast or should I simply abandon them and try to write my own minimal RX-like text matching routines?