Here's a few test shots I took today comparing the field of view of the 15mm fisheye (both with and without software correction) and the 17-40. Exposure was set to manual and held constant through the shots, white balance was set the same for all shots, and I used a tripod. The camera is a 5D.
Image 1: 15mm fisheye uncorrected
 | HTTP response: 404 | MIME changed to 'image/gif' | Byte size: ZERO | PHOTOBUCKET ERROR IMAGE |
Image 2: Image 1 after being corrected in DxO
| HTTP response: 404 | MIME changed to 'image/gif' | Byte size: ZERO | PHOTOBUCKET ERROR IMAGE |
Image 3: Image 2 cropped in Lightroom to match Image 4
| HTTP response: 404 | MIME changed to 'image/gif' | Byte size: ZERO | PHOTOBUCKET ERROR IMAGE |
Image 4: 17-40 set to 17mm
| HTTP response: 404 | MIME changed to 'image/gif' | Byte size: ZERO | PHOTOBUCKET ERROR IMAGE |
Image 5: 17-40 set to 24mm for reference
| HTTP response: 404 | MIME changed to 'image/gif' | Byte size: ZERO | PHOTOBUCKET ERROR IMAGE |
Overall, I like the fisheye. As you can see in the first picture, it doesn't have the stereotypical fisheye look except near the edges. The two trees look bent inward and the utility pole in the extreme upper left corner is definitely leaning.
Once corrected, you're still left with a very wide angle image but now the parts of the image close to the edges are stretched. The tree on the right and the shed on the left really show this.
Cropping to match the field of view of a 17mm lens cuts out the most distorted parts of the image. I had to crop out 30% of each dimension, reducing the pixel count by 50%, to create image #3 from #2. This is not a problem for me but if you plan on making big prints, that's a lot of pixels to toss out.
