Kind of surprising how soon people forget what has happened before... prior to Apr 2003, there were reports of the hammer being stuck at 800 MHz and having lower IPC than the K7 and the fanboys predicting the death of AMD.... you all saw what happened later.
The K8 had all the strengths of the K7, but it also had many of its shortcomings (poor performing L2 cache, poor integer performance, etc) Against the P7 core, it looked wonderful however, but its shortcoming were revealed with the introduction of the Pentium-m that gave the mobile hammer a great fight. The K8 didn't really evolve, the P-M core got tweaked a lot and morphed from competitive to leading (banias -> dothan -> yonah -> merom/conroe) The introduction of yonah was the prelude, as we saw the P-M surpassing the K8 in almost everything except media, and the speculation that if that got fixed it would be the fastest CPU existing... it got fixed, and many of you have the result of "that fix" powering your rigs.
Back into the subject, there is no way the K10 will not be have at least >30% higher IPC than a K8. Most of those shortcoming in the K8 (several carried over from the K7) are being taken care of. AMD is in a better situation than it was 4 years ago. Of the "CPU business", Jerry Sanders always focused more on the "CPU" part, while Hector made AMD really get into the "business" part..... AMD still has who is arguably the brightest mind in the processor world, Dirk Meyer. Hector preferred to give stability to the business and make money, get return of the investment instead of keep pushing faster and faster CPUs very often. They even bought ATI. AMD had 4 years to improve K8, and if many ignorant fanboys saw intel back on top, it is highly unlikely Hector Ruiz or Dirk Meyer didn't They knew intel would counter the K8 eventually, but they probably didn't expect the answer to be that good. Now it is their time to come back to the CPU part and show the work of those 4 years.
My prediction
K10 15% faster clock-for-clock than penryn in single theaded integer apps.
K10 20% faster clock-for-clock than penryn in single theaded floating point apps.
Add 5-20% to that amount in multithreaded apps, and loads with high amounts of data exchange will be higher. AMD still has a more advanced architecture (integrated memory controller and HT)
Barcelona will beat clovertown in <2U servers, and wipe the floor with it in >8u configurations. Phenom will beat penryn in the desktop. How much?