Drew McLellan writes about the downside of incremental upgrades to web applications:

"Do everything you can, no matter how sound your reasoning, to group process and interface changes together into a package. Call it a new version or whatever and tell the user about the changes you’ve made. Be aware that every change you make forces the user to relearn and costs them time, no matter how sensible the change. Be aware that if a process can be refined to save the user a couple of seconds, yet it takes them two minutes to relearn, it's going to take sixty times through that process before the user sees any time benefit."

I would argue that the bigger your user base, the more conservative you have to become, which is pretty much how eBay for instance evolves its user interface at this point. But when you're still ramping up full speed, the benefit of delivering a better experience outstrips the relearning hurdle inflicted on "incumbent" users. In other words there's a life cycle for web applications too, which doesn't mean you can't reset the clock with major releases once in a while. Blogger 2004 comes to mind as a successful example of that, much more convincing than the Dano update last year.

09/07/04 update: speaking of eBay redesigns: Product Strategy: Making eBay Easier and More Intuitive.