We need to analyze this with more precision. Because, the 3 ms you are getting for rendering one of your GUIs (if it is one of the big ones) is not a bad number. Anyway, HUDs should render faster.

Could you test with/without MSAA/PPAA and at different resolutions? With this we can identify the bottlenecks.

Sorry for my late reply. I tried different resolutions and anti-aliasing modes and didn't find any differences in rendering speed. So I'm sure it depends only on amount of controls displayed.
When I enable the NoesisGUI camera, in the Unity profiler appears "Gfx.WaitForPresent" call which is usually means that CPU waiting for GPU rendering (VSync off). I uploaded a screenshot.

If there is no difference in speed when changing resolution or multisampling mode, then the bottleneck is probably located in the communication CPU <-> GPU. This was greatly improved in v1.2

Could you link a screenshot of the GUI that corresponds to that profiler stats?

Yes, this link: http://screencloud.net/v/qgK7
On this UI only one opacity group used (for "version (Alpha)" control in the top middle). All other controls has been re-done to use opacity of the brush.

We can wait for 1.2 beta for PC/Mac/Linux: the game is still fast enough now even when rendering very complex controls.

Are the icons in the left corner vector images? One of the things we want to add in future releases (not 1.2.0) is support for BitmapCache. That way all those vectors get rendered and cached as bitmap images boosting the performance (although... increasing the memory usage probably).

Are the icons in the left corner vector images? One of the things we want to add in future releases (not 1.2.0) is support for BitmapCache. That way all those vectors get rendered and cached as bitmap images boosting the performance (although... increasing the memory usage probably).

Yes, they're a vector images... I knew about it and replaced most complex elements with bitmaps (such as "grid" under radar control), but I didn't get any noticeable effect on render speed, maybe because we have so many other vector paths...
The BitmapCache is a great feature! Especially if we will be able to configure it like in WPF.

Yeah, BitmapCache would definitely be amazing to have.