Since I got a small collection of IR filters now, I gave them a test in the park the other day.
Test kit: Canon 450D with all filters removed. So no IR block and I think the AA is gone too. Camera is therefore sensitive to IR, and probably somewhat to UV too although not likely to a significant degree. Lens is EF-S 15-85, which I only realised later on meant I couldn't test another filter I had with me at the time.
Settings: Aperture priority f/3.5 and auto-ISO. Manual focus using magnified live view on the trees in the middle.
Processing: Picasa used to raw convert. Auto-colour + Auto-contrast. Resize for web in Photoshop Elements.
Not the most inspiring scene, but I chose this since there were various tree colours. This shot is without any filter in front of the lens. Picasa had a go at correcting the large amount of red from IR, and it looks kinda photo-realistic but the colours are still a bit off.
The infamous Hoya R72 filter (720nm). Note the sky is slightly reddish, and the foliage is slightly blueish.
Cheapo ebay Neewar branded 850nm. The colour hues are largely gone. Hotspot effect in middle.
Cheapo ebay Neewar branded 950nm. The colour hues are largely gone. Hotspot effect in middle. Slightly softer than the 850nm image.
Can we explain these results? Using the chart on this
page for the 40D (assuming the 450D is similar) then I think we can. Compare the RGB channel responses past 720nm or so. Of particular interest is the blue channel which is only significantly sensitive beyond 790nm. So foliage reflects relatively more of that. The sky is reddish, suggesting it has more of the shorter wavelengths.
Overall, what can we say? A 720nm filter would give you the potential to do some interesting false colour images. A common trick used for IR landscapes is to swap the red and blue channels, giving you a blue sky while retaining white-ish subjects. If you don't want the false colour effect then the longer wavelength filters could help there. The 950nm cuts out a lot of light though, so exposures may get tricky. Also note that diffraction effects become visible sooner the longer the wavelength is.