Perhaps what sisandra is using as test routines doesn't use the additional SSE3 extensions by nature? I'm sure there is documentation of this somewhere online (forums probably atm).
As for Prescott, Go for the Northwood and skip Prescott.
For those who care for my usual lengthy explanation, let's reflect back on the overall situation.
All AMD & Intel fanboys step down for a second, this is a game of leapfrog not underdog.
AMD's recent history shows a lot of success in taking existing architecture (x86) and building upon it, for instance their first release of the Athlon was quite powerful compared to the p3. Coincidentally enough (or not) this happened about the time that DEC sold off their 'Alpha' cpu technology and most of the engineering staff who built that 64bit cpu (which was VERY advanced and powerful for its time) subsequently left for other companies. AMD, as you might guess, hired a few of these excellent & experienced minds. The even pioneered the SSE style accellerations by implementing 3dNow!, which Intel basically copied & extended to create SSE/1.
Not only did the original Athlon fare well against the p3 but Intel's 'next generation' architecture in the first few P4's to market lost to both the Athlon *and* their own 1Ghz+ P3's. Intel was attempting for the first time to bring some of their research from the 'high end' Itanium (or Itanic as some non-fans prefer to call it) and in the beginning of this they let it be known that while initially the technology seemed tepid it should allow them to scale nicely as the P4 clocks up. They *also* let it be known that their first P4 socket, the 'socket 423' would be soon replaced by 'socket 478' which should run for a long time after. So many people ran out and bought the P4 and were disappointed with what they recieved compared to their peers who bought AMD Athlons.
Once the P4 passed the 2Ghz mark it started pulling ahead of Athlons in many tasks (except 87fpu, ie, older plugins). This was especially the case due to memory throughput. By the time the P4 was getting over to 2.4Ghz 'hump' (which required a die-shrink) AMD countered with their 'Athlon XP' line which brought a lot of improvements (full SSE/1 support and smaller size/cooler) and again AMD brought cheers from the 'little man'.
Now we are nearing 'modern times' (relatively speaking) and Intel has not only had to rev the core and give their FSB (front side bus) a boost to 800mhz to keep their latest P4 core (the northwood) up with AMD's Althlon XP, but AMD has managed a sidestep and brought 64bits to the desktop with an even newer core found in the Opteron/Fx/Amd64 line. They also integrated a memory controller and added support for SSE/2 (which Intel added to the original P4). Most of the performance boost that they have gotten in modern applications actually stems from the integrated memory controller's effect on memory access and the integration of SSE/2 and other 'enhancements' (larger caches etc). The 64bit 'extensions' affect ONLY how much total RAM can be addressed by the chip (ie, accessed and should not be confused with the precisions of calculations taking place in the cpu (32/64 or 80bit (x87fpu is actually 80bit)).
Intel has been a bit slower in bringing the Prescott to market than they would have liked, because in order to get the next die shrink for it they have had to introduce a few new things. First they had to pioneer a 'strained silicon' process to reduce the problem of electrons leaking into nearby pathways, so that they can keep operating voltage within reasonable engineering levels. They are also introducing a new socket yet again (and there are probably other 'features' present in the new cores which will not be enabled until they can be 'fixed' and enabled in later generations to improve performance).
Now with where we are NOW, Prescott is 'almost' here and AMD has had since last summer to improve their Opteron/Fx/AMD64 core. AMD performs quite admirably when compared with Intel across the entire x86 board (Opteron vs. Xeon, Fx vs. P4EE, AMD64 vs. P4) and Intel's next generation seems to be a losing deal compared to not only AMD but also their own current P4's!
Sense that we've come full circle?
The 'Prescott' core should once again 'scale quite nicely' over the next year or year and a half (not to mention that it has 1Mb L2 cache), and once it hits the 'sweet spot' of its performance curve I'm sure Intel is confidant that it will perform up to par with the AMD stuff. It will be interesting to see what happens too because Intel is competing with AMD's dual advantage of not just 64 bit extensions (probably already in Prescott but disabled) but also AMD's integrated memory controller which Intel has no current plans for. However where Intel REALLY has their work cut out for them is with the upcoming socket change. "Socket T" uses a ****** type of connection, which means that rather than the cpu having pins that extend down from it, it simply has very VERY small bumps rather like ball bearings and the socket has very tiny spring loaded pins which extend up to contact the socket. This allows Intel to advance the overall 'process' which is used to create their cpu's and reduce the amount of copper between the cpu itself and the 'packaging' which the small core sits upon (the green plate in modern AMD & Intel cpu's, (Athlons, durons and P3's were brown).
Understand that the 'spring loaded pins' which extend up from the socket are a potential achilles heel as it may be you can only change cpu's a few times before affecting the springs & contacts to the point where the motherboard becomes useless or buggy. This doesn't affect people who buy a system and upgrade it once at the most, but it is possible that people who love to 'tweak' their hardware will revolt to AMD's side en masse if this turns out to be the case. Or it may just be that buying a new motherboard becomes part of that 'upgrading' and tweaking. Or it may be that the engineering problems aren't as bad as it would seem. Noone knows yet because Intel certainly isn't telling.
Now to summarize the last 2 paragraphs real quick, basically Intel is going to introduce the Prescott at a performance disadvantage on socket 478 and move it to Socket T, a platform change that may have a potential achilles heel. It may not however, and Socket T motherboards will have advantages: initally many boards will support PCI Express, and eventually both DDR-2 and a 1066Mhz bus will emerge as well as universal support of PCI Express and a total abandonment of the ISA bus (which still exists to support the serial/parallel & ps/2 connectors). If the Socket T proves to be up to the task, then by the time that the Prescott is at 1066Mhz the 4.8Ghz speed should overcome the deep pipeline and low IPC issues.
Now as for AMD, at the time that they announced their 'Hammer' core (used in Opteron/Fx/64) they also announced a 4th variation: dual core cpu's. So about the time Intel's Prescott peaks AMD may have a tasty treat of their own and bring true multiprocessor multithreading to the masses. Of course Intel will have Nehalem coming soon after and they have hinted that what we now know as HT (hypertransport) may evolve along similar lines as well.
In the long run the IA/32 platform, even with AMD's 64bit extended lease on life may follow Intel's plan and IA/64 (Itanium) or something like it may trickle down to the masses. IA/32 has other shortcomings that the 64bit extensions don't address (such as IRQ!) and unlike AMD Intel really has the R&D financing to play a 10-year game with that one. We'll see....
Incidentally I may be off a few degrees here & there...feel free to correct me
(don't get angry, I'm probably drunk... 
)