Of all the things that could impact the performance of a tool like ffmpeg, the operating system probably matters the least. Much more important is just the hardware you are using: the processor(s), the hard disk and the RAM. I use Vector Linux, but any distribution will work fine. You can even use windows if you really want to.

OK, I think I found something that sort of works. The code for the script is here:

http://dark-code.bulix.org/kdqxmw-67863 (with syntax highlighting and line numbers)
http://dark-code.bulix.org/kdqxmw-67863?raw (without syntax highlighting and line numbers)

Before you can run the script you need to patch your ffmpeg with the following:

http://dark-code.bulix.org/x6iy8f-67862?raw

which is an ever so slightly modified version of the patch on this page:

http://lists.mplayerhq.hu/pipermail/ffmpeg-devel/2006-October/017666.html

After patching the ffmpeg source tree (make sure the resulting pip.c file ends up in the vhook directory), you will need to find the line in your Makefile that looks something like this:

BASEHOOKS = fish null watermark

and change it to:

BASEHOOKS = pip fish null watermark

Then, before the script will run, you need to take all the videos you want to concatenate together and turn them into .mpg files with the same dimensions and frame rate. Then, you need to open up any image editor and create an all black image with the same dimensions as your videos. The script will print out usage information if you execute it with no arguments. Here is an example of how to run the script:

$ ls
video1.mpg
video2.mpg
video3.mpg
all_black_320x240.png
fade.sh
$ OUTPUT_FILE="final_video.mpg" BLACK_PICTURE_FILE="all_black_320x240.png" FADE_FRAMES="25" ./fade.sh video1.mpg video2.mpg video3.mpg

And that's it. Please note that this script doesn't exactly "cross-fade" the videos... the transition is "fade to black," but it might work for some things. As far as I can tell a true cross fade is not possible. I know this is not pretty, please someone improve it :-)

I had thought that this was not possible (imlib2 only allows changing the opacity of text), although I think I might have found a solution for you. I'll try it out later today and let you know if it actually works :)

How many geek points do I need before I can trade them in for some new hardware?

The non-linear editor I use is cinelerra-cv. It works great. Although if you are going to be expanding still frames into movies (like I had to do with the "highlighting") I would do the expanding in ffmpeg. Otherwise you'll spend unnecessarily large amounts of time zooming in and dragging then zooming out and dragging some more etc. etc. to get the still frame expanded to the right length. The highlighting is just making two still frame movies (one highlighted and one not) and ramping the opacity between them.

For screen captures I usually use pyvnc2swf and then jump through some hoops to get it formated in a way that can be combined with other video clips. If you don't want to deal with that you can use recordmydesktop, which is also very good and much easier to use. Some people like xvidcap but it is not functional at all on my box (1.46Ghz Celeron Laptop.), it needs much more juice than my machine can pump out.