I further expanded on my previous Koch Snowflake Processing sketch to allow for dynamic zooming and moving. This Dynamic Snowflake is also at OpenProcessing.
To deal with memory issues for huge numbers of points, it only draws down to the level of 2px-long edges. Thus, only so many pixels are visible at a given time. Additionally, the zoom factor has an upper limit so that it does not increase indefinitely (but you can still keep zooming).
I rewrote nearly the entire program to implement these changes, and found the previous Edge class and heavy use of recursion not practical for the problems I was trying to solve. This surprised me given the highly recursive and self-similar nature of a fractal, and I think these projects were valuable case studies for me in algorithm choice.