Jens,
I don't agree that all copyrighted images should be deleted. I don't want to delete the GFDL images I've uploaded, and I assure you that I have not placed those images into the public domain, so the Wikipedia must either accept that it can use them under my GFDL licensing of them or find some other alternative (using as fair use, or not using or some other option).
However, it's not that simple either. Consider http://en.wikipedia.org/wiki/Image:Devonport_Dockyard_in_1909_plan_small.png . That's an image I released as GFDL and I'm perfectly entitled to do so because I'm in the US. However, there's a potential catch. The source image is in the public domain in the US (an old British book) but in some jurisdictions that book may still be covered by copyright and the image may not be in the public domain. Then there's the matter that I obtained the image from a web site which had scanned the book. In US law, that's fine - the book is in the public domain, so the scan is also in the public domain. But what about the law elsewhere? I don't know that there's no jurisdiction where that would be copyright infringement, even though I know that it wasn't copyright infringement where I am and that I was perfectly entitled to make and release a GFDL derivative work. Somewhere, there may be a jurisdiction where that work isn't actually GFDL. So I fully described the origins and anyone can check their local laws
The point here is that even GFDL works do not have clear status. Any prudent reuser must check the copyright status in their own jurisdiction before using them.
Now, it could be argued that someone could take a current photograph of the same subject. But what about the different attitudes to rights to the appearance of architectural works around the world? Again, even if the image had been taken by the uploader, a reuser must still check what their local law says.
This sort of question is one reason why publishers often ask contributors of anything which will be published to agree to indemnify the publisher against legal claims. The contributor may be ignorant of the law in Middle Of Nowhere, but the publisher at least gets to sue the contributor to recover damages, if the insurance piolicy doesn't pay. The contributor, of course, is likely to be unhappy if they didn't know that the work would be published in Middle Of Nowhere and didn't include that place in the thousands of different jurisdictions around the world they checked before uploading the image.
Again, the message here is: you always have to check. It really doesn't matter what you claim the license is, whether you're claiming public domain so no license needed or GFDL license. You have to check.
It doesn't matter what we accept, and since we're a wiki there's no reason to reject temporarily less than ideal images, because they can be replaced later. What matters is that we get accurate descriptions and tag things to help people to identify possible risks, so they can consider things under their local law and make an informed decision for themselves.
The images from secretlondon/Caroline are another example. Caroline correctly obtained licenses. Those protect her and the Wikipedia from allegations of copyright infringement in the jurisdiction from which she was uploading. In the US, those images are fair use in the way we use them, so I'll shortly have added a fair use tag to them as well. Same image. Which it is depends on where you are. Pick the legal status which makes most sense for you. For US Wikipedia users, it'll be fair use and they can use the image that way. UK users and others who may need a license knowwho to ask if their local laws require that they do so.
The world is a comlex place. Tagging and telling people origins makes more sense than forgetting that we're a wiki and can replace images later, if we can get a more free image. That's how we handle text stubs. It should also be how we handle image stubs (images with less than ideal rights status). The images are not the last version we'll ever see just because they are all we have today.