That makes sense when the client is always a known quantity, but such is not the case with the web. While you can get an idea of what the majority of your target audience is running, there's always going to be some percentage that will be left out when writing client code for the majority browser only. There's always going to be client situations that you can't predict much less write for, and in those cases more server-heavy solutions tend to do better.

Is a traditional desktop client really a better-known quantity than a web browser these days?