Rik
I sort of thought that Sandy had covered this, but I notice some inputs which I think make clarification of your question and their answers important. Firstly much depends on the scanner you are using (and you say you are using an Epson, me too). Some scanners (not ours) have the ability to adjust lamp light intensity and the analog gain of the system to enable the scanner operator to tune the device to get the useful dark and useful light areas of the film you intend to scan.
In contrast to this the Epsons can't adjust much (well I think there is a sort of short cut into that as I've written on my blog
here however I've also noticed that similar effects can be had by forcing the blue channel slider all the way to the black in colour work too). So when you scan something you will probably only get a portion of what you need in the range of the scanner. This varies from film to film. The Epsons seem to be optimised around Slide (which is not a bad thing). For example, with black and white negative I get a result like this:
There isn't much need for the stuff on either end, so this will be truncated by adjusting the level sliders (the little black and white triangles left and right of the histogram).
Now, this can be done in Photoshop or it can be done in the scanner driver software. Photoshop provides some useful tools to set the clipping points visually so people may just leave it were it is and touch up in Photoshop.
This is where having only 8 bit scans will run you aground, as you will get some posterisation or banding occuring (especially if the range you have on your neg (like if it was a C-41 chromogenic neg) is not as wide as this one.
Now, if you were to trim up in the scanner driver (and the scanner and driver is smart) it should be smart enough to do this in 16 bits internally then dish up an 8 bit result with greater precision.
I think now is a good time to mention that 8 bits is a string of 8 digits which can be either a 1 or a 0 (right?) so the number 256 (in base 10) requires 9 bits but if you don't mind counting 0 as a value then 8 bits will work for representing 256 levels (or 0 to 255). So when you 'strech' the histogram you are not going to get neat values on subtle tonal graduations. Essentially think of this as:
if you divide 5 by 2 you'll get either 2 or 3 but not 2.5 (as it won't exist). So you will need the extra precision of 16 bits when adjusting your raw scan to avoid this as it can then interpolate a value
Now, its when you work with colour negatve that you have your biggest challenge. You may not have looked, but if you do a linear positive scan of your colour negative you may see something like this
notice how little blue there is in there? So you are unavoidably going to be stretching that when you scan. If you don't have one of the above mentioned scanners (with the capacity to tune your sampling) then you're gonna get banding and noise in your scans (pepper grain with that anyone?). Now like it or not it is exactly this data which your scanner will have to work with, so whether you use 'home mode' or 'professional mode' and tell the scanner that you are using colour negative (instead of positive) the software will work with the same thing as you see in a linear positive scan. The difference is that you're (hopefully) more intuitive and programmable than the machine is, so by scanning in positive and inverting, assigning profiles, adjusting, tweaking yourself you'll get a better result (which is why you should
never send a machine to do a humans work).
Well, I don't want to really dribble on here for ever, so I'll leave it here for now, hopefully you can see that while it is possible to use 8 bit scans well (if you have a drum scanner) it is better to tune your scans a little in the software, then tune them in photo editing software in 16 bits. At this point you can certainly save space by saving in 8 bits (as minor tweaks will be unlikely to produce banding).
From here you can work in 8 bits (with all your major levels / curves adjustments done) especially as you say that some of your preferred filters only work in 8 bits.
If you read those links it'll perhaps help you understand more (and since you're an Epson user I recommend
this page of mine too if you do colour negative).
