I have been playing with a patch to the 2.6.37-rc kernel that speeds up compute and memory intensive applications. Its call the Transparent Hugepage Support patch written by Andrea Arcangeli. Basically it works by reducing the number of TLB entries a large application needs. It also increases the effective size of memory that the TLB cache can cover. This results in a fewer TLB cache misses which require TLB look ups to service.
A more technical description can be found here
and in the patch's kernel documentation here
The only benchmark I have had time to do right now is a build of xulrunner. I observed a speed up of 1.4% (about 23 seconds on a 27 minute compile). Andrea Arcangeli has a patch to gcc to make gcc faster still but I haven't tried that yet.
I can say Firefox and Gnome in general felt a bit more lively but I don't have any benchmarks to back it up.
It is supposed to really speed up databases and virtual clients a lot.
Sadly I don't have enough disk space to run the Phoronix Benchmark.
The patch can be found here