quality is not equal to pixel count!
This is exactly right.
A large pixel count increases sharpness (which is desireable) but it also increases the visibility of the noise (which is typically undesireable).
Obviously reasonable looking transfers occur at some nominal pixel count. But increasing the pixel count, beyond some point, starts to make the transfer look noisy (but no more noisier than the original film). If you have a bad transfer due to too many pixels you can do a low pass filtering of the signal either by:
a. downsampling the transfer to a lower res
b. or doing another transfer at the same lower res.
As long as both produce equivalent results they are, by definition, equivalent.
There are alternatives to conventional low pass filtering (ie. alternatives to the above options) which seek to maintain sharpness of the signal while suppressing the noise. For example, the "Reduce Noise" filter in Photoshop CS5 (which is not a super-res algorithm by the way) uses an interesting application of wavelet theory to infer the signal and subtract the noise based on that inference. Not everyone's cup of tea. Some prefer the noise over such.
ALL approaches to noise reduction (including low pass filtering: such as lower res sampling) introduce "artifacts". The only question is which algorithm/physical process satisfys one's personal artistic judgement.
For some, it is not a question of reducing noise at all. It is a question of reproducing the original film (including it's noise), as best as possible. For those who prefer this then it is simply a matter of using the highest number of pixels (and highest resolving transfer lens).
And Roger - a technical test, using a high-def scan, is just a test. The argument you were maintaining, that because you've done a high-def scan there is no point doing a lower def scan misreads the test as if it were a recommendation for day-to-day activity. But a test is a one-off act. If the test suggests you could have used a lower def scan, then that is knowledge you can usefully apply to future scans. It is obviously too late to use that knowledge on the original test scan. You obviously can't go back in time and get your money back on the hi-def scan. But who, (apart from Kent perhaps) would argue otherwise?
However I know your argument(s) are probably better than I've charcaterised them here, and you've elaborated some better with Chris (eg. why does Chis do HD scans if he's on record as saying PAL SD is all you need).
I don't want to pre-empt Chris' reply too much but I'd argue that it is easier to get an optimum SD result downsampled from an HD scan than it is to get the same result from a physical "SD" scan. Downsampling allows you to tweak the filtering on a computer (which is a pleasant experience for many). But a physical SD scan requires you tweek the filtering through adjustment of lenses and sensors (which for me, at least, is often an unpleasant experience). Or if not using your own rig, then how much do you trust that an SD transfer house actually does a good SD transfer? It is more likely that an HD transfer house has put a bit of extra effort in to their setup if they want to maintain their claim as an HD house.
On the other hand I'm not actually aware of what, precisely, Chris said. There is a world of difference between a "typical" SD scan an "ideal" SD scan. And your argument could very well be a very good argument if "typical" is what Chris actually said/meant.
Hi Kent - did you manage to get to this sentence without interpreting everything else I've said here as "blah blah blah"?