No, no they definitely ran themselves into the ground.
That they have.
They wasted time pursuing MCM quads when they should have strapped two x2's together, like Intel did for the q6600.
You are contradicting yourself. AMD did not release any MCMs until well after the Core 2 Quads were obsolete (2010), and that is only for high-end server configurations, with 8 or more cores. Intel's Core 2 Quad, OTOH, was an MCM, and AMD's choice of single-die integration gave them an edge in many multithreaded situations, that could hide their single-thread performance deficit v. Core 2. They may have made quite a few bad choices, but that was one of their better ones.
Then they chose to delay delay delay bringing Bulldozer to market on a new process (32nm) when they should have shrunk their Thuban x6's to 32nm first, and then worried about Bulldozer.
If Llano (a more important 'shrink') and BD are any indication, a shrunk X6 wouldn't be too much better than a 45nm one, except at low speeds, and it still would be a hindrance going forward. That AMD was able to use a modified K7 to compete with up to early Nehalems is, on one hand, an indication that it was done right the first time. However, that they kept using it over and over, with only minor changes after the Hammer, most of it away from the FUs, is an indication that there were improvements not pursued, and/or that they should have been doing much more aggressive R&D on future technology, and weren't
(signs generally point to both, for the same reason: R&D costs money that could be better used for golden parachutes and insider trades).
If anything, they should have made the new arch able to perform well at lower speeds, and/or have much improved performance/Watt, regardless of what speeds they wanted to be able to reach. Even if they could have gotten it faster at launch, or it can get much faster (IOW, cooler) over revisions, what good would that be when their halo parts were still only going to be ~2-2.5GHz, and future desktop/mobile versions will need to perform well at such lower speeds, too?
Then they pissed away all this money on Bulldozer and didn't bother working with Microsoft to get the thread scheduler set straight to work as it does with hyperthreading.
Intel didn't either, and it took awhile for Hyperthreading to work well on Windows. HT just happens to be old, now. MS has also had many minor updates to fix scheduling problems for certain configurations, and there is no sign that's going to stop. IMO, the blame for that is all on MS: they should really make a scheduler that can handle shared memory (caches) well, with API/ABI compatibility layers where needed. This sort of thing is only going to get worse as time goes on.
Note, also, that while BD's performance in Linux, which has a better general scheduler, may be superior to Windows at this time, it still only beats Phenom IIs doing things most of us don't care about, uses plenty of power to do it, and single-thread performance is still every bit as bad as in Windows.
(did I meet my proverbs quota for the day?)
Encore, encore!