It's interesting to see the two different sides of this debate. Is 0.8 the problem because it doesn't conform to the established rules of the system or are the older clients the problem because they use a crappy database which can't handle large blocks?
From what I can gather it seems to me like the database used by 0.8 is much more superior, so it will be used. But in order to patch 0.8 it seems like an artificial cap will be placed on the block size so that 0.8.1 is compatible with older versions.
The bug is that there has been an unknown BDB limit the in 0.7 and earlier code for quite some time. It wasn't known about, so therefore measures weren't added to avoid provoking the bug in the older code.
0.8 is at "fault" for not being perfectly bug compatable.
0.7 is actually at fault for being broken..
The <= 0.7 problem is not "fixed". The mining pool rollback just got everyone on the same blockchain again and bought some time to patch the older code. The chain won't fork as long as > 51% of the hash pool is on 0.7 but that won't stop smaller operators generating valid blocks that cause trouble in the future.
The bdb bug has to be understood and fixed.