David, I've used both methods, and they both have good and not-so-good aspects.
The in-camera noise reduction seems to me to do a more complete job, but it also doubles the amount of time it takes to capture an image. That eats up your time, and also eats up more battery charge, both of which limit the field experience. The in-camera method also seems to miss a number of hot pixels, which have to be fixed in Photoshop anyway using the clone stamp tool.
Making a single dark exposure and then subtracting it in Photoshop is useful since you only have to make that dark exposure once every few frames. This can save a lot of field time, but also requires you to fully register the image and dark frames when doing the subtraction in Photoshop (using the "difference" blend mode on a dark frame layer). If you don't get the images precisely registered, you only make the noise rough and don't remove it completely. But be careful not to go too long between frames for that next dark frame. If you are using the camera continuously, the sensor can self-heat a little, changing its temperature (that's where the battery charge is going when the battery drains - dissipated away as heat in the electronics). The noise field changes as the chip temperature changes. You should make a dark exposure at the end of the night when it is all warmed up as well as earlier when things are cool, if you are shooting continuously through the period.
With my Canon 20D, I've noticed that neither method does a complete job of removing hot pixels, even though the same pixels on my sensor seem to be hot each time. That requires a deft hand with the clone stamp tool at high magnification in Photoshop.
- Kit