Yeah it's removing the dust and scratches . . .
In terms of speed and quality, Coolscan 9000 + Nikonscan ICE has no equal when it comes to it's magical ability to remove dust and scratches. Here is an example of a particularly dusty and scratched 35mm Kodak 160VC film scanned using a DSLR and the Coolscan with and without ICE.
Coolscan setup has normal and fine ICE modes, takes about 30 seconds to scan with ICE off and about 50 seconds with ICE. For 35mm, I have autobellows for my DSLR and once setup I can take a copy in seconds - but of course it has no ICE and color negatives will require inversion. The post process it would take to render a positive alone would already take far longer then a scan and still not be as good. I didn't bother cleaning it up because I highly doubt I could match ICE regardless of how long I worked on it.
However, if you only shot true b&w or slides, the post work would be much easier but stitching - shooting and post work, would be tedious and add quite a bit of time.
Although I fully understand the convenience of the Coolscan 9000 for dust and scratch correction, I need to make a few remarks here.
First off, the 9000's ICE certainly has merits, but it also has visible effects on the image rendition, in particular on sharpness and smoothing, which make it more useful as a productivity tool to get adequate quality in a comparatively short time, than as a tool to get great images - which is perfectly normal, since the dust specks and the scratches remove information that cannot really be recovered. All the software can do is try to fill the missing parts of the image in an unconspicuous way, but it does not magically make them come back. It also needs to avoid spoiling the parts of the image where there is no defect. According to
this example you gave in another thread, even the 9000 does not completely succeed there, especially in the fine setting. In other words, I strongly disagree that ICE does not do anything else than removing dust and scratches, that would be too good to be true and the signal theory says otherwise. Whether or not you can live with these artefacts for the sake of saved cleaning time, is entirely a matter of taste. Personally I don't like this look and I'm willing to spend a few seconds on each image to avoid it, but to each his own.
A big advantage of the DSLR approach is that it (usually) uses diffuse light which, unlike the pointlight used in scanners, does not emphasize the grain, dust and scratches. This makes software removal easier and less destructive.
As for the presence of defects on the film, there is no alternative to carefully cleaning it. Use an air blower as a first attempt. If not sufficient, use a fine brush, antistatic if possible: it only takes a few seconds, less than ICE processing time, and it is more effective at cleaning the image - read what I said above. If still dirty, you can also resort to wet cleaning with PEC12 or pure alcohol - but NOT water as I've seen: water makes gelatin swell and get sticky, which is a great way to cause mechanical damage, attract more dust and make it very difficult to remove. There are solutions for scratches, such as fluid mounting (works for both DSLR and scanning) but they are not very convenient and certainly reduce productivity if it is a priority for you.
As regards negative to positive conversion, I don't do much colour negative myself but I have had good success (and good colours) with the Photoshop curves approach as long as there are white and black parts, even tiny ones, in the image. It takes a little time, but no more than scanning it in good quality. You can also do it with Vuescan, by just importing the raw file from a DSLR and processing it as if it was a scanned image. These are just two solutions that work, among others.
Finally, and although it is probably stretching a bit the topic of the discussion, it is probably good to point out here that it is only the Coolscan 9000 that can do ICE on any kind of film. Any other scanner (including other Coolscans) will not be able to do ICE on non-chromogenic film, aka B&W and Kodachrome. I find this reminder particularly useful given the scarce supply of Coolscan 9000's, their cost on the used market, and the fact that they are virtually unserviceable now. I wouldn't want anyone reading this thread to believe that this is a "scanner vs. DSLR" instead of a "Coolscan 9000 vs. DSLR" discussion, and be disppointed becaused they bought another scanner due of a misunderstanding. [Edit: oh, I hadn't seen you had said it before. Sorry, I'm being redundant on this one]
Incidentally, in this example of yours, that I linked to above, I notice that even without ICE your D800 is very noticeably sharper than your 9000 (granted, in 135). It would be interesting, since you seem to have tested the resolution of all the configurations you have used, to show the comparison between the 9000 and the D800.
