Midtskogen's trick is pretty simple and powerful with a good choice of blur size. Another technique you could try is to characterize the light pollution and remove it by subtraction or difference modes in an image editor. Here's a quick summary of how I do it using GIMP:
1) Open the image
2) Choose the "color picker tool", with the settings "sample average" over a large area (a good portion of the image area), and "set to foreground color". Then click on the image over an area that is relatively empty of galaxies/nebulae/stars (mostly background noise). The idea is that you're detecting the average color of just the light pollution in the image, which in a deep sky image should be pretty uniform.
3) Create a new layer filled with this foreground color. Set the layer mode to "difference", and then slide the opacity until it best cancels out the light pollution. I find this is usually around 50%.
This is what I get using this method + a bit of curves adjustment on your ISO800 image. Scaled up 200% to see effects a bit better:
This is a really
nice capture by the way. After the little bit of processing you can even see the star formation regions!
As for the question of ISO vs. exposure time, always go with the lowest ISO / longest exposure time combination that you can manage while obtaining a neutrally exposed image. Higher ISO can achieve a neutrally exposed image more quickly, but it will be more noisy.
The theory is that you want to maximize the signal to noise ratio (S/N), which represents how much of the light being recorded is actually from the object you want to image, vs. noise from random fluctuations. S/N increases as the square root of the exposure time. Raising the ISO itself does not do anything for S/N -- it's just acting as an amplifier of both signal and
noise by binning pixels together. But with higher ISO and shorter exposure time, you'll get a noisier image.
Other ways to improve S/N are to stack several images together, which acts to average out the noise.