When looking at the loopback TCP network performance by transferring 10GB over the loopback interface, the performance was uninteresting with the 2.6.32, 2.6.33, and 2.6.34 kernel releases. However, with the Linux 2.6.35 kernel, the result was interesting and it is certainly an improvement. The loopback TCP network performance was faster by about 17%, which may be attributed to Google's contribution of transparent spreading of incoming network traffic load across CPU cores via Receive Packet Steering and Receive Flow Steering, along with other network optimizations.
Lastly, the performance of Open FMM Nero2D turned ever so slightly for the worse with the Linux 2.6.35 kernel. With the three previous kernels, the time to run this test of a parallel electromagnetic solver was 685~688 seconds, but with the just-released Linux kernel the time was increased to 700 seconds.
If you are considering the upgrade to the Linux 2.6.35 kernel and are running higher-end (Intel Core i7-type) hardware, it looks like there are some performance gains to appreciate in this release compared to the older 2.6.32-ish kernels found in Ubuntu 10.04 LTS and other distributions from early this year. Our daily kernel benchmarks hasn't found any significant performance changes from the 2.6.34 cycle to the 2.6.35 final release, but we'll continue to run a couple more benchmarks in different hardware configurations to look for any other performance changes with this newest Linux kernel. Your own performance tests or thoughts on the Linux 2.6.35 kernel release can be shared in the Phoronix Forums. Besides these couple of performance gains and losses, the 2.6.35 kernel provides more Btrfs improvements, Intel H.264 VA-API decoding support for Clarkdale IGPs, and ATI Radeon KMS power management support, among many other changes.