From the moment they released the first version of the kernel, it had broken error reporting (would just crash without reporting) and had bad lockup problems. This was reported early and ignored through the alpha, beta and release cycles. The obviously unreportable and undiagnosable lockup problem was "patched" by users (for the most part) by switching to the RT kernel.
Various serious problems were reported and closed out without being fixed and reopening them was not allowed, e.g. breaking Broadcom 43xx wireless cards.
The analysis at the time that made the most sense to me was that taking a Debian unstable snapshot every 6 months and then furiously trying to make it stable enough was impractical. My interpretation of the above development management problems was that shipping on schedule took absolute priority, and the whole process reached a tipping point with this release.
Me, I eventually went to Fedora, then to Debian stable lenny because I needed Xen and stability ... and by then the hardware I had bought for this was old enough that Debian lenny supported everything ^_^.
(I can supply plenty of URLs or my entire set of notes if desired.)