Gisli,
I've actually been considering the same thing, but have yet to try it in the field. I'm still waiting for the ice to thaw from the lakes around here 
I think what you want to do will be very challenging to do underwater. Most panoramas are built with the assumption that that the camera will be rotated around the nodal point, but that won't work underwater. Instead you have to change perspective by moving from point to point and snapping pictures.
For your question, I don't think that lens speed (f#) is a big issue as you don't need to high shutter speeds to freeze a fast moving subject. I think that you're wreck will accommodate slow shutter speeds, so the biggest issue is minimizing the water column between you and your subject. IE, you need to get as close as you can and still get as much of the Wreck into the viewfinder as possible. For this, fisheyes are king. Especially in murky waters.
I wouldn't worry too much about the distortions that are introduced by it's use. These can be corrected for in the panorama generating process. Your biggest issue is going to be in the execution of collecting the images. I think large changes in scale (moving closer and further from your subject), small rotations about image axis (horizon not horizontal) and changes in the scene due to lighting variations are going to be the biggest issues in getting a good panorama.
Here is a link to a tutorial that outlines how to do it using open source (free) tools. I'm planning on doing a couple of dry runs on the surface to get more familiar with the system before I try to do it underwater. You might want to do the same.
http://www.dojoe.net/tutorials/linear-pano/
Good luck!
Mike