Ok I admit I wasn't in love with XAML at the time when it was launched, mostly because of its verbosity and bloatedness :) But about a month ago I had to implement a kind of complex design into a .NET desktop application, and it was that kind of design that could have required lots of weeks to develop using owner draw controls and GDI+: list views with clipped contents, multiple-step gradient borders etc, they all proved very easy to do in XAML like a few hours for the most complicated control.
As for WPF being bloated, some time ago I thought of implementing a hardware accelerated DirectX-based XAML rendering engine, yet I found it already is GPU accelerated:
DirectX 10 cards are able to cache the font glyphs in video memory,
then perform the composition (assembling of character glyphs in the
correct order, with the correct spacing), alpha-blending (application
of anti-aliasing), and RGB blending (ClearType's sub-pixel color
calculations), entirely in hardware. This means that only the original
glyphs need to be stored in video memory once per font (Microsoft
estimates that this would require 2 MB of video memory per font), and
other operations such as the display of anti-aliased text on top of
other graphics – including video – can also be done with no computation
effort on the part of the CPU.
Still I kind of doubt the renderer is using the GPU at it's full power - I can bet the gradients are not done using shaders and the shapes are not drawn with polys etc.