I don't understand this. There's a reasonably clear HIG document in GNOME which is continuously updated. The strict packing rules make the resulting interface flexible in regards to RTL languages and reasonably consistent with other applications. GTK+ has widgets for just about every purpose. Since the introduction of Cairo it is trivial to draw custom widgets, especially using the gtkmm and cairomm C++ bindings: the custom tab bar could be done in a day. With those bindings it is even quite easy to modify the behaviour of existing widgets by overloading their signal handlers, Murray Cumming and the others working on those have invested a great deal of effort to truly (and easily) expose the object system via C++.
And as a useful side-effect, those pesky testicles will shrink and get out of the way.
I agree with you that this issue seems to have been thoroughly ignored by the release engineers, who pushed the 2.6 driver through despite its downfalls. (apparently to gain support for some newer chipsets, which could've been simply backported by adding PCI ID's to the 2.4 driver)
Compositing and 3D performance are horrid and even non-composited 2D is slow when EXA is used. Apparently, if you're lucky you can switch to UXA, but it is non-functional on my systems. (G35: no modes found & i915: DRI disabled)
The issue can be improved (but not fixed) by using a backport of the version 2.4 intel driver available from the "http://ppa.launchpad.net/siretart/ppa/ubuntu" PPA of siretart. This provides minor performance enhancements but is still slower than intrepid.
With all the fancy scientists in the world, why can't they just once build a nuclear balm?