But here's why I think you might be wrong. Your argument rests an one assumption, which is that the big datasets, the gigabyte size video you're editing, the huge science data set that you want to analyse, etc, originate on your PC.
That used to be the case traditionally and it continues to be the case quite frequently. But I think it may be the exception in a few years time. Video cameras could simply buffer the recordings and stream it right back to the data center. Satellite pictures and other kinds of scientific data could also be stored directly from the sensor to the data center instead of being stored on some PC and then uploaded for sharing purposes. Other types of data that is being analysed today is already stored on the backend and cannot be downloaded as a whole (RFID data, financial trading kind of stuff, etc). It has to be manipulated and analysed on the server.
If most data lives in google's data center the equation changes. It might be impossible to download the entire dataset and then edit it on your PC using a desktop app. The bandwidth argument suddenly favors the browser. High powered desktop computers could be relegated to a few special purpose tasks. Most users might not own machines capable of manipulating massive datasets or video files.
And then you have things like google's native client. It's not like everything client-side has to be written in JavaScript forever. There's not much left that a browser equipped with something like native client cannot do provided the data lives in the cloud.