That google video is great!
So in watching that I think I've come to a couple of conclusions.
The first one is that it might be best to crunch my numbers back on the Arduino. My original thinking in sending out raw data is that I'm more comfortable with programming on the computer (Max/MSP) than I am in Arduino, and it's much easier to test/adapt/modify in that environment. I also thought it would be faster since my computer CPU is significantly faster than an Arduino.
I didn't account for the bottleneck of serial communication through. So all my fusion stuff will happen potentially displaced by serial latency (unless I'm incorrect in thinking that data travelling across a serial pipe is slower than data in a void loop circling around and around).
That's a big change from what I'm doing now, and I'm in the middle of a project based around this so I'm not gonna 'jump ship' just yet (and just use the 6axis data I have).
It also looks like quaternions is in fact what I'm after (and what that code does).
Now it's just a matter of how much of that code can live in Arduino land (and how much is based on their hardware/protocol).
The final thing is that I'm wondering if the way I have my sensors oriented is a problem. You can see my boards here in an old photo when I had first put them together.
I have the magnetometer directly above the MCU. It is pointing in the correct direction, but I don't know if having something directly below it (a couple millis as there's poofy double-sided tape) is a problem.