One of my twins (see sig) has been having a problem. After a time (varies) he starts to break down. Example: random program craches, segmentation faults, lock-ups. It requires me to hard boot to rescue partitions on another (safe) drive and fix the ext2 file system on the primary disks partitions for the main system (/ and home, not the usr) I must also delete several things from the /tmp to kill the leftover locks.
These twins are identical except for the harddrives.... Each was running patched 2.4.26 kernels straight from the Kernel Archives with various patchs required by my system...

With nothing to lose I used the generic 2.6.7 kernel on the troubled twin... The results where... well remarkable.
So I compiled the 2.6.7 to the box and there is a marked increase in performance. This box now runs as fast as its twin (running a 2.4.26 kern). I have lost the openMosix set-up which is ok as I have been running PVM shells (very interesting, several things run via a PVM shell across several machines) which does not require patched (same) kernel versions to run.

So now I will be getting the other twin up to this kernel level and I have finally got a 2.6 level kernel to work. (if you where not aware I failed several times to build a 2.6 kernel).
I still have the Dragon for openMosix experiments...

Now I must also research the use of 2.6 kernels in cluster applications, !!!fun stuff!!!

It is still to early to tell if I will have the same problems with this twin using a 2.6 kern over a 2.4 kern.
I noticed better hardware id over the 2.4 level and seems to be more stable so far. I was able to "fix" som module issues that had started happening with the 2.4 and I did not need to patch i2c for the SMbus and sensors.

I now feel 2.6 level kernels are worth the effort...

Well, normal desktop performance will be better on a 2.6 kernel. But when I did a single heavy task (game) it wasn't running as fast and smooth as it does on a 2.4 kernel. Besides that, I'm still feeling the 2.6 kernel is a development kernel, so I returned to a 2.4 kernel. Problems with alsa drivers, a reiserfs panic and lm_sensors problems were reasons for me to return to 2.4. I was using 2.6.8.1 .

aha.....my biggest problem with 2.6 kernel is the udev... at this moment my video camera & modem still not working with udev...

Nichole_knc also remember that there's an option in the kernel configuration to select if you want to give more priority to current thread (for desktops) or same for all (servers).

The twin still has a problem though not as bad as with a 2.4 kernel.

2.6 handles lmsensors just fine here, no patching required.
Alsa is a pain no matter what kernel you build.. I believe in to be very backwards that you much un-install/re-install the drivers (most common fix) when you make a small change in the kernel or add other modules.
I agree with the "development" statement related to the 2.6.# kernel. The higher the number the higher in the "development" tree it is. However you can argue this about a 2.4.## kernel as it is also in a state of constant upgrade in the tree (read development).
The system is faster over all not just in a GUI.
As for the udev thingy just turn it off if you don't want it and have/had a working /dev directory before udev took over.
IT is real easy to do..
In a console as su/root. `chmod 0 /etc/rc.d/rc.udev
then `reboot`
Once the system reboots it will return to the "regular" /dev if you had one properly set prior to udev all will be well. NOTE you may have to fix some premissions.
OR you can be edit the configure files for udev located in /etc/udev so that the missing hardware is loaded.
I tried devfs before udev is supposedly the replacement.... Ask me THEY are both a pain...

I am not sure your problems are related to software, instead I would check the hardware, in particular the memory (there is a nice tool called memtest to do that) Is the processor run hot ? You could test the temperature with proper lm_sensor module for your motherboard.

Memory has been tested and replaced as well as shuffled...
Yes the processor is cool. I have 2 thermal coupling attached to each one.
The processor is running at 91.4F. The internel case temp is a cool 79.7F.
The other twin (compiling) is running a 93.2F with a case temp of 86.6F.
I too believe it hardware related but I am puzzled as to why the hard drives fragment out as bad as they do only on this box...