I think your frustration is due to a little misunderstanding. If the inability to install the nvidia driver is cause for this dissatisfaction, then rest assured that it's not the packager's fault. Installing kernel modules means placing the files in the correct directory.

A real-time kernel, or any kernel on a distro _will_ have some sort of "extraver" or "localver". It's the extra characters after the actual linux version, sometimes/some distros including the -rc or upstream patch version. Confusing? No worries. Examples galore:

$ uname -r
2.6.25-rt-sidux

$uname -r
2.6.25.4-rt21

Whatever it is, you have to make sure the nvidia stuff go into whatever the correct -rt kernel directory is in /lib/modules ;) The following are the steps similar to what a particular distro buildscript takes:

bash NVIDIA-Linux*pkg0.run --extract-only
cd NVIDIA-Linux*pkg0
cd usr/src/nv/
ln -s Makefile.kbuild Makefile
make SYSSRC=/lib/modules/$thelibmoduledirasshownabove/build module
install -m644 nvidia.ko /lib/modules/$asabove/kernel/drivers/video/
depmod -v $asbove
rmmod nvidia
CTRL+ALT+BACKSPACE

Maybe there's a command you can use try run a help on the proprietary executable. Then, there are other packages like ndiswrapper - and oh yeah catalyst who can forget that one - that may need a little hack or two on the rt-patched kernel. A checklist:

1) Upstream kernel
2) Upstream kernel patchset
3) RT kernel patchset
4) Apply whatever hacks needed (so far only catalyst/fglrx needs one)
5) Install proprietary module to correct directory
6) Good luck and enjoy!

As for performance, I've always been having better with older kernels as newer ones roll out. I'm currently on 2.6.23.1-rt11 :) Realtime performance actually is no longer a problem. It used to be a problem sometime ago, but after migration to pam security and all that it's become a breeze. On Windows recording with my laptop's Intel HDA is a no-go @ > 500ms latency + ASIO-4-All, but on Linux I get 20+-.