CFS? I was talking about BFS... was that a typo?
Anyway, BFS was written for desktop computers with only one or few cores and it is said to work better than the default cpu scheduler in these systems, but it works worst than the default in computers with lots of cores. I won't try to explain in detail as it is better reading this: http://ck.kolivas.org/patches/bfs/bfs-faq.txt

Before the BFS, Con Kolivas wrote de Staircase CPU Scheduler, witch I used to compile a custom kernel for Diablo, if you have a N800 or N810 you can download it here: https://garage.maemo.org/frs/?group_id=1200

Unfortunately, I'm an electric engineering student, not a performance products specialist, so I'm not familiar with benchmarking performance. But I can try describing my experience with the Staircase CPU Scheduler in my N810.
I was trying to make the N810 a better multitasking device, but it had a 128MB RAM + 128MB swap limitation in it's default configuration. So I first tried increasing the swap space by moving (all) the system files to the internal flash card and formatting the /dev/mtdblock4 device as swap. That didn't work well because the system would take too much time to move data from ram to swap a from swap to ram, making the device unresponsive, and the watchdog would reboot it. So I compiled the patched kernel and not only the multitasking experience was faster, but it was also stable, the reboots were gone because the Staircase CPU scheduler wouldn't let the system become unresponsive anymore. That's the goal of the schedulers Con Kolivas wrote: make the system more responsive increasing the end-user experience.
So, that's a lot of talking but I won't be able to provide you test results to prove what I'm saying, but I hope I convinced you to try it yourself.
Sorry for my English, I hope you could understand it.