And of course when this is shown to be problematic with a simple (GPL v2) license, the answer is to add tons of complications to the license to the point where you can't say that if you read both licenses literally, that the GPL v3 and BSD licenses are compatible.....
I actually kinda like the GPL v2. The GPL v3 license is an abomination however and needs to die.
Before you jump on me about BSD license compatibility, let me add a caveat and explanation. The caveat is that although the texts of the licenses are at least likely incompatible, "everyone" (namely every free software-inclined lawyer I have asked about this issue) agrees that since the intent was to be compatible the GPL v3 should be read to be compatible, but no two lawyers give me the same answers. I have discussed this with both Eben Moglen and Richard Fontana however. Also one cannot dispute that the GPL v3 is clearly compatible with the MIT license since that explicitly allows sublicensing.
The problem has to do with section 7 additional permissions and additional restrictions in the GPL v3, as well as the lack of a sublicensing grant in any of the BSD licenses. The BSD licenses offer a public grant by the software author to all who obtain the source code. However, they allow other intermingled code to be subject to other licensing schemes. This is different from the MIT license which directly allows sublicensing (Author A publishes, I take his work, and change the effective license without changing the software, and sell it to you, offering only some of the rights I received). The GPL v3 arguably requires a sublicensing grant from a permissive license in order to be compatible and the BSD family does not give this sort of license.
So when I talked with Moglen he suggested it would be compatible because you could add whatever additional restrictions you wanted but those would be unenforcible until actual changes were made to the software. Since the license change would be unenforcible, he suggested, the original author would not have his or her copyrights infringed. Richard Fontana however has suggested that this is a bad assumption and that developers must not make such an assumption that this is safe.
Instead Fontana suggests that additional notices should be read broadly enough to include the BSD license (and license grant) as is. However, this guts the idea of additional permissions under the GPL v3 and the requirement that they be legally removable without altering the code. Since you can't reduce the BSD license to (GPL v3 + permissions that can be removed without altering the software), the license is not compatible with the text of the GPL v3.
This means that although the licenses are (everyone agrees) compatible, it isn't at all clear as to what this means. Nor is it clear if I write a BSD license which explicitly states that sublicensing is not permitted and that permissions can only be revoked upon altering the code, whether this would also be compatible.
In short the solution to a license which may have some corner cases is a license that lawyers can argue about endlessly.