Parallel generate was already using multiple processors by spanning multiple processes in NX 7.5, you just had to set a customer default. With NX 8.5 this was improved for automatic separation, the customer default is still needed to specify the maximum number of processors/cores to use.
The whats new guide and the Siemens PLM Americas connection should include a slide about the time savings when generating cavity milling, z-level profile and some other operation types.
ISV has a big performance improvement in NX 10
3D verify has a big performance improvement in NX 10
If you check the what's new guide of each NX release, you should get plenty of information about performance improvements