Thanks for linking this Jan - somehow I missed this blog post - but it is a very good idea. I'd be interested, though, to know whether there is much impact on the client's experience on the page - as this approach is pushing all of the work to the client's browser as opposed to changing the actual images ourselves. I'd just be careful because this operation could get expensive (read: slow) if there are a lot of images on the page.

I agree, it is a workaround for a missing solution.

Did anyone log the requirement for an image path per skin into Gemini?

I was hoping they would send two copies of the current Gemini installation (each spinning the oppersite directions) down the Large Hadron Collide in the hope of logging a Higgs Boson issue :).