This guy dissects the dhtml techniques that make google maps so kick-ass:
http://jgwebber.blogspot.com/2005/02/mapping-google.html
Google Maps ought to be a target scenario for anyone working on a UI platform. All you have to do is answer these questions (which flash sometimes does really well):
1. What goals did the developer want to achieve?
2. What obstacles did the developer work around in order to achieve those goals?
3. How do you address those obstacles in your platform?
Answer to #1:
The goals are:
1. Build an interactive application.
2. Visualize dynamic data.
3. Use open and popular technologies to avoid 3rd-party dependencies.
Answer to #2:
The main obstacle is that there are no open or mainstream standards that can realistically achieve #1 and #2. Page refreshes are not interactive. It's arguable that neither pdf, svg, flash player, swf, or applets meet the combined requirements of openness, popularity, and dependency-safety.
DHTML meets goals 1, 2, and 3 but not without some tenacious hacking around obstacles that had driven many such as myself to Flash.
Problem: Browsers make multiple streamed data-sources difficult to aggregate.
Solution: IFrames, tiled images, server-generated images
Problem: Browsers make 3D effects difficult.
Solution: DHTML z-index with 8-bit alpha PNGs
Problem: Browsers can't render graphics at run-time.
Solution: Server-generated graphics and run-time positioning/layer with dhtml.
Problem: Support frequent queries on a massive scale
Solution: Build a search-engine company.
Answer to #3:
Maybe question 3 is the wrong question, obviated by question 1. Question 4 = "How often is dhtml good enough?".