I don't know why the flock elements would be getting further and further away, but I believe "flocks" was designed specifically so that the elements would not collide. Flocks are tricky to work with in terms of controlling precise movement.
Springs is also something that I believe currently isn't quite working (fully implemented or supported) in current versions of A:M. Or more precisely, I've never been successful in recent versions of A:M of getting springs to work.
What acts similar to "flocks" and what you are describing might be done with applying "translate to" and "orient like" constraints with lag (of 1-3 frames).
This animation test (done in 2009) uses constraints with lag. Not quite what you are describing, but parts of each model are constrained with lag to follow other parts of its model.
Or even perhaps what you need is something using a dynamic constraint?
Or perhaps some sort of Newton simulation?
Without seeing an example, it sounds like you are describing something that is done with Flash, where elements follow the cursor? or links/nodes presented graphically look like they are held together with "strings"
What is a Newton Simulation?
On another note you got me thinking in a new direction. Would it be possible to create a model with a center main bone and lots of other spheres (20-60) with child bones that could be positioned by expressions that are related to the parent bones rate of rotations?