Should I port SLAM to the Raspberry Pi's unused GPU on my Autonomous Rover?

Right now, I use Arduinos for motor control (PID control for encoders and dc motor) as well as aggregating sensor data and feeding the Raspberry Pi on my Autonomous Rover.

The Pi runs a live stream webcam for the rover, manages the webpage control console, and maps the environment for dead reckoning navigation.  This is a lot of work for one little CPU.

I've recently seen a few articles on using the dormant GPU on the Pi. and though that Simultaneous Localization and Mapping (SLAM) would be perfectly suited for the GPU's capabilities. Real world mapping and localization should be practically identical to Video Game based algorithms... It's been almost a year since Broadcom opened up the specs for this chip.

I would also use the GPU for calculating shortest path via A* routines. 

So... has anyone looked into this? 

 (And yes... I "Googled" it first...)

References:

Andrew Holme: Accelerating Fourier transforms using the GPU
Pete Warden: How to optimize Raspberry Pi code using its GPU
GPGPU hacking on the Pi
Hacking The GPU For Fun And Profit (Pt. 1) 
Broadcom released the specs for the VideoCore IV GPU
SLAM: Remotely-Processed Visual SLAM Using Open-Source Software