Getting back to why I started this thread...
I guess I'm just too stubborn or cheap to upgrade to the 24-105 L. I sent my 28-135 off to Canon and had it repaired. It's still the same "good but not great" 28-135 I recall. At least it's symmetrical again instead of the left side being blurred.
Really I'm quite satisfied with the center of this lens, and on my Rebel XT I don't hesitate to use it at any aperture. Purchasing a 5D revealed it's shortcomings, though: the edges are not great, and the corners are pretty dismal wide open. So I decided to try the free evaluation copy of DxO Optics Pro 3.5 for a 5D with 28-135 lens.
I had test frames of a bookshelf from when the lens came back, so I ran five of the wide open frames, each at a different focal length, through DxO. I settled on using 28mm for a fair test since I would have to correct some obvious barrel distortion in Photoshop. DxO would be correcting it for me. I left all of the DxO adjustments at default/auto and had it output to DNG. I then set the white balance the same between the DxO and non-DxO images, turned off sharpening and noise reduction in ACR (set to 0), corrected vignetting in the non-DxO file in ACR, and opened the files in Photoshop. The non-DxO file I then corrected the barrel distortion on (Filter>Distort>Lens Correction) and cropped it back to rectangular. Then both files had Photokit Sharpener capture sharpening applied, then output sharpening for inkjet glossy at 240 PPI (which the files were sized at to print about 12x18 inches), and printed.
Well, I had to repeat the procedure three more times, for the non-DxO frames shot at f/4.0, f/5.6, and f/8.0. Looking strictly at the prints, the center of the frame on the DxO-processed f/3.5 print is definitely much better than the f/4.0 non-DxO file, and really darn close to being as good as the f/5.6 non-DxO center. I'm really splitting hairs here, they're almost indistinguishable (DxO f/3.5 versus non-DxO f/5.6). But here's the kicker: where the 28-135 really has problems, the edges and corners, the DxO processed file looks as good at f/3.5 as the non-DxO processed file that was shot at f/8.0! The DxO-processed f/3.5 file is actually a bit better than the non-DxO f/8 file in the extreme corner. I'm surprised... shocked even.
I know, I should post some crops on the web so everyone can gawk at them, but I'll leave such antics to Michael and the other dedicated reviewers. Feel free to try it yourself for free. I did, and I'll almost certainly be purchasing DxO myself.
Disclaimer: I'm in no way affiliated with DxO, Canon, Michael, or anyone else you've ever heard of. These are just my personal observations.