I use XNA 3.1 at the moment with premultiplied alpha (native Premultiplied Alpha support was added in 4.0).
Mostly it saves me time going through and making sure the transparent areas of my images have the correct colours (to avoid this kind of thing). I also port to Silverlight, which does premultiplication by default.
XNA has a content pipeline which makes adding premultplied alpha preprocessing super-easy.
That link contains the code for doing the conversion. For every pixel:
c.R = (byte)(c.R * c.A / 255);
c.G = (byte)(c.G * c.A / 255);
c.B = (byte)(c.B * c.A / 255);
There is a lot to be said for having a content pipeline. If you're not using XNA, but are using Visual Studio, then you could throw together one yourself. Any other IDEs/build processes should also be capable.
And finally, if you don't have a content processing step in your build (and so odds are you don't have a lot of content anyway), you could just as easily perform the premultiplication at runtime when you load your textures. It will add some constant factor to your texture load time.
(If the implication wasn't clear: you probably shouldn't be editing premultiplied alpha images. Export them premultiplied if you can, or better yet preprocess them.)