New SSE instruction set with the Penryn models.
SSE4 to be more precise. Only beneficial if you do video encoding using a codec that is optimized for SSE4 instructions. DivX 6.7 or 6.8 has an experimental SSE4 mode, and I think encoding performance boost is about 30% over not using SSE4.
High k hafnium gate architecture is able to switch faster than previous 65nm counterparts.
In other words compared to a similarly clocked (CPU speed) Conroe, the Penryn has been tested to perform about 0% - 10% better depending on the application. I think gaming was on average 5% or 6% improvement. The highest was with video encoding for about 10%.UPDATE
Part of that increased performance can also be attributed to the larger cache.