Multisampling / super-sampling

To remove Polygon jaggies (not texture anti-aliasing) is sio2 setup to have the scene rendered as an image larger than screen (say 125%) and then display a scaled down version of that image with anti-aliasing. Other solutions?

Multisampling is not available on the device by default however you could always do it in software since the color buffer can easily be manipulated...

does this mean the PowerVR GPU does not offser any form of antialiasing?
pretty astonishing...

Basically what is on the device is not 100% PowerVR, not all the features are included only a subset.

"you could always do it in software since the color buffer can easily be manipulated..."
Any hints as to how to manipulate the color buffer?

Is it a case of doubling the viewport size (glViewport( _x, _y, _w*2, _h*2), rendering the scene and using glCopyTexImage2D or glCopyTexSubImage2D to copy the color buffer and then mipmap (or average the values) and render resulting texture to screen?.

What is the likely performance hit? My frame rate is 40 on device with doubled viewport, am I likely to drop below 30fps when copying color buffers about.

I think the option to software antialias would be a great addition to the engine if anyone has code.