> Compression is about as CPU intensive as you can get with audio outside of voice recognition
Highly polyphonic physical modeling synthesis is more CPU intensive than compression... then you start adding on many instances of complicated reverb effects, some effects that require fft/ifft passes, autotune, &on and on, all with realtime constraints (not to mention all of the copying of buffers and context-swapping that goes on with the way these systems are designed)... and you get a high-cpu monster pretty quickly.
Highly polyphonic physical modeling synthesis is more CPU intensive than compression... then you start adding on many instances of complicated reverb effects, some effects that require fft/ifft passes, autotune, &on and on, all with realtime constraints (not to mention all of the copying of buffers and context-swapping that goes on with the way these systems are designed)... and you get a high-cpu monster pretty quickly.