The discussion on this answer to the question \"How can I use Google\'s new imageless button?\" Has prompted this question.
Google seems to think that going imageless i
The page load is smoother as no images have to be loaded and will appear later than the rest
The button text is also readable in the case somebody cannot read/view images, yet you have the graphical look. (building a graphical button with images in the traditional way around real text is as complicated HTML as this method)
As they mentioned in their blog, these buttons are skinnable without creating and storing custom images.
Basically, you get all the advantages of plain text buttons over custom imaging, while still having a nice, skinnable graphical look.
Btw, the "imageless" button might as well contain an image inside the visual template. This approach is quite similar to XAML's approach to templating and styling the visual tree.
Isn't this done because the height of the button may vary (for example the text size)?
I think in this specific case I can only see the advantage that the buttons can be programatically generated. If you don't know what your button will say it's probably easy to make this way than generating it using somekind of image library generator.
Also changing one CSS can make you change the look-and-feel of all buttons at once. Using image buttons you'll need to update everyone and each of images.