The question is very vague. Can you be little specific? When you say data intensive, is it more data coming from and going into the database or you have less updates and more selects? An example would be better.

.

.

Three things need to be understood, how the data will be captured, what are the processing requirement, like how frequently they will be processed, how it will be consumed? And again in each part there will be other questions.

Parallel programming with Hadoop

A desktop client would be ideal for data intensive application. For example a WPF/Silverlight application. As the local system memory can be used for streaming / loading of data desktop application performs better compared to a web application.