You’ve selected a nice – difficult – problem domain. You could speed up the whole simulation by implementing a faster nearest-neighbor algorithm, reducing your search time from Order n^2 to Order nlog(n). Consider Quad Trees, which aren’t that difficult to implement. The Voronoi suggestion might be too much complication for your needs. … Do you have some bugs? There are quirks when bubbles disappear, grow in discrete increments, merge, etc. Give some more consideration to the size of the resulting bubble when bubbles merge.

It would be cool if it were interactive — you know how you can draw on a condensated mirror with your finger? That should be pretty easy to simulate with the mouse repelling the droplets.

The perfect circles takes away from the organic feel.

I agree with a lot of the other comments: I think the piece would be more successful if you applied faster algorithms and varied the shapes of the spheres. It is obvious you did a lot of research for the project though–good job! I also admire your determination in choosing a difficult problem (and not playing it safe). Here is the wikipedia side on the nearest-neighbor search (the nlogn algorithm is described here):
http://en.wikipedia.org/wiki/Nearest_neighbor_search
-Amanda

Vornoi Cells:
http://en.wikipedia.org/wiki/Voronoi_diagram

interesting subject for a simulation, it sort of feels like its asking for another extension or a new form, like adding interactiveness or whatnot. could be cool if you use it to create type.