Are you using a ROI to limit the amount of processed pixels to a minimum?
How many channels are you processing?
With one proc that could be what you will have to face using such extreme settings.
Looks like the DirBlur doesn't like putting up with just one cpu, the differences are amazing.
Here is a quick benchmark test applying a 165 pixel dirBlur (radial) to a 500x500 pixel roi on a 2k square frame with 16 channels:
1 cpu: 12 minutes 39 sec
2 cpus: 1 minute 38 sec
3 cpus: 1 minute 5 sec
4 cpus: 49 sec
this is on a dual core opteron 285 - 2.61 GHz (windows xp)