The short answer, yes you could divide the image (tiles) and distribute then to each computer over a network. The fact is, there is no script/function to do that and the overhead for each connection will be something to consider, since will need to send the whole scene anyway to allow the render engine show all reflection, color bleeding, etc...
Another approach, that I am testing right now, is to render a image with a smaller number of samples with different seed for each computer and then, re-assembly them back. This methodology is already covered by others (Kent Trammell, this is the arquive file on cgcokkie or even here @StackExchange.
We can think about this as: For two images generated with different seeds, let's get pixel[10,10] for instance. On first image the RGB value is (0.34, 0.78, 0.2) and on second image the same pixel[10,10] is (0.32, 0.80, 0.22). None of this values are right or wrong, but we can assume that the best value (or median) will be (0.33, 0.79, 0.21), that is probably more close to right answer for that pixel.
Using this approach, is it possible to write a simple script to load images from a folder, and them find the median value for each pixel and generate a final image.