Before you have a chance to become confused, you may wish to consider the following:
Dean's first paragraph is irrelevant because we are not discussing scanner profiling. Scanner profiling is a "good thing to do" for dealing with reflective scans and positive transparency scans for the same reason that any device profiling is worthwhile, and that advice is valid regardless of what image editing options you elect to perform. Your question wasn't about profiling the scanner, so let's move on from there.
Dean's second paragraph is correct, except for scanner-dependent operations such as lamp control (in Nikon scanners) and infra-red scratch and dust removal on colour transparent media (positive and negative) using scanners and applications that support it. Unless the scanner has user-adjustable lamp-control, exposure adjustment is not a scanner dependent operation. It is software controlled. For dust and scratch removal, both SilverFast and Vuescan support it for a range of scanners. You can check on the SilverFast website for the scanners that allow for infrared detection; SilverFast's iSRD tool is much more granular than the infrared detection tool offered in Vuescan, but both can do it and removing all this debris at the scan stage is a vastly superior solution to scratch and dust removal in Photoshop or any raw converter.
Dean's third paragraph is correct, but you knew that and you are asking where it is better to make these adjustments.
Dean's fourth paragraph can be correct, depending on the kind of scan you produce. If you produce an adjusted scan in the scanning software and export it to your hard-drive he's right. All the adjustments done in the scanning software are baked-in. That is either OK or not so OK, depending on the quality of the adjustments and how permanently you want to keep them.
SilverFast and Vuescan both offer the facility to make a "raw" scan (in SilverFast called "HDR"). These scans are not "raw" in the same sense as a raw data file from a digital camera. These scans are fully rendered, three channel pixel-based images by the time you see them (and that is true in Vuescan too, whether they have a "DNG" container or not). They are only raw in the sense that nothing else is done to them except to dump everything read by the scanner's sensor into a processed, linear, but UNedited image file, so nothing has been destroyed that you have any control over. You can keep this as a master file and never damage it by never saving over it. For example, you can call it up in Photoshop, make any adjustments you want, then do a "Save As", the original file you were working on goes back to sleep as it was - unaltered and undamaged, while the newly created file has your adjustments.
The other way of doing this "non-destructively", which is where Dean's point is correct, is to open the scanned image in Lightroom or Camera Raw and make your edits there. Nothing happens to the file until you export it, if you ever do, because all that happens within the raw converter is that a metadata instruction set with all your edits sits with or in the image file, and only when you do something such as printing it or exporting it for whatever purpose, those edits get implemented on the fly ON THE EXPORTED IMAGE, so the original remains unaltered/undamaged, but the exported image has been altered.
Turning to Dean's final paragraph, there can of course be differences between Dean's preferences, factual considerations and what suits you. The factual consideration is that each digital image editing application only has so much "adjustability" built into its controls. So the more you can safely edit within the scanning software, even though it is baked in unless you use the raw scan option I defined above, the better the starting point of the image for the next level of adjusting you may wish to do in another image editing application, and the less further editing will be required. This is logically unambiguous and proves itself in reality too, especially for very difficult images.
One reason why you may wish to deploy software outside the scanning application for image editing is simply because it can often do stuff that you can't do in the scanning software because the scanning software wasn't designed for it. For example, no scanning software allows you to correct pin cushion or barrel distortion or to implement perspective controls, whereas Lightroom/ACR do. Another reason to consider the options between the scanning software and say ACR/LR, is that much of what you should do between these applications depends on which tool is best for the job at hand, and that varies; so my basic point, and what I emphasize in my book, is not to be dogmatic about this. There is NOT "one size fits all". SilverFast, for example, has a white balancing tool that is quite unique, insofar as it lets you balance four separate zones in one image where each one needs different balancing, at once, with four mouse clicks. It can work really well, and I haven't seen another application that does this. Lightroom I would contend is much stronger than ANY scanner application in teasing out shadow detail while maintaining and even enhancing the quality of deep shadow tones. Don't look for uni-dimensional recipes in this work. It's horses for courses. Some things work better in some places than others. Have the options available, experiment, and use whatever gives you a more satisfactory final result.