On some StackOverflow question about GWT cons and pros from 2009 someone answered that:
[…] Use gwt to only fill in the dynamic parts of your apps. So if you have some complex user interactions with lots of fields. However, don’t use the panels that come with it. Take your existing stock designer supplied pages. Carve out the areas that will contain the controls for your app. Attach these controls to the page within onModuleLoad(). This way you can use the standard pages from your designer and also do all the styling outside the gwt.
Well, I’m trying to figure the reason behind it. If I understand correctly, the client receives a compiled HTML and JavaScript files. So, from the browser perspective, isn’t he supposed to be oblivious to the way the HTML page was created (statically by editing the HTML page or dynamically by GWT code)? Also, it appears that the GWT panels are translated into simple HTML elements on the compiled page. So where is the performance hit here?
Try to imagine that your app have 5000 different complex unique pages. If you try to implement it all using GWT, you will have quite a big problem during development, because compilation time will be very long and devmode will be slow as well. Sometimes it is better to implement everything in GWT, sometimes it is better to combine static pages with GWT.
Basically you have to use GWT when it will bring some real benefits to the user. If user won’t be able to see difference, if page is static or dynamic, use the technology which is easiest to use for you. For example if you are developing web image editor, GWT is an obvious choice. If you are creating some Wikipedia clone, you should use GWT mostly for some dynamic parts, like administration, search and editing.