Publishing quality full-color material, whether in book or magazine form, is one of those things that’s so fiendishly difficult to do right that it’s nearly impossible to explain why to a layman. There are so many ways for things to go wrong and no way at all for them all to go right; even the best work ends up being a compromise.
Mangetout has given you the basics, and link he provided has a reasonably good description of the printing process. If you’re talking about work of the quality you see in National Geographic or your typical coffee-table book, you’re talking about material that’s being printed at a 175- to 250-line screen frequency, on extremely high-quality (and expensive) coated paper stock. Only certain presses are able to hold that fine a screen, and then only with very experienced and skilled pressmen. Those presses tend to be ones that run several hundred impressions a minute (at least), and you can generally count on several stop-start iterations once the job is set up on press and the pressmen attempt to tweak the ink densities and other press settings to get the appearance of the page to match the contract proof. That’s just to get one side of the sheet right; you’ll be doing the same thing over again on the other side. I have personally attended a press check for a job where we ran 30,000 sheets to get the ink densities set exactly right on the first side of a double-sided single-page job with a total press run of 1,000 – by the time you turn the press on and off again, another 500 sheets are through. And that was for a job at 175 lpi on a medium-high quality magazine stock.
The film used to make the plates has to be extremely carefully calibrated and linearized, with appropriate allowances made for the press gain characteristics of the particular press that’s going to be printing the job. It also has to created from an imagesetter with extremely fine tolerances and, for screen frequencies this high, one of the better halftoning algorithms. Processor chemistry and timing have to perfect on both the film itself and the platemaking process.
The scans of the photos have to be extremely sharp and packed with dynamic range, which means that they almost definitely have to be done on a high-end drum scanner using photo-multiplier tube scanning elements instead of the less expensive charge-coupled-device elements used in almost all flatbed scanners. Even with extremely high-quality scans, the results have to be carefully color-corrected, and unsharp masking and various other image manipulations applied in order to achieve the desired results.
This assumes, of course, that you’re starting from high-quality, medium-to-large format color transparencies, shot on professional-quality film; at the very least, the photos will need to be on good-quality 35mm slides (and probably shot on film no faster than ISO 400, preferably slower, for an acceptably low level of grain), and the focus and exposure will need to be spot-on (meaning that we’re assuming no distortion or other artifacts of sub-standard optics in the camera body or lenses). Prints are extremely forgiving of a wide range of exposures (meaning your local one-hour photo guys can almost always make acceptable prints from your print film, no matter how badly you botch the exposure), but slides and transparencies give you very little leeway. And of course, all the way back at the beginning, we assume that the lighting and composition of the photos themselves is first-rate.
For computer-generated images (or for scans, they’re not that much different once they’re pixels), you can expect to need to generate the images at a resolution equal to approximately 1.5 times the screen frequency you’re targeting (the exact values will depend to some degree on the halftoning algorithm used – I’m not going to get into rational vs. stochastic screening here) – assuming that the images are being generated at the size they’ll be printed.
As you can well imagine, most of these steps are frightfully expensive, especially when work has to be redone or tweaked to get the desired level of quality. Even with months of lead time to allow for doing the printing overseas (generally in the Far East, where the labor cost is much lower and there’s a reasonably well-developed fine printing industry), the costs are high (particularly if you have fly to overseas for several press checks). This type of printing is only cost-effective for fairly large press runs (tens, if not hundreds, of thousands of impressions). Lots of really smart people have tried to come up with something that produces high quality pages that’s cost-effective at lower run lengths, but so far no one has – there’s always a tradeoff in quality. The Heidelberg DI presses and similar beasts have gotten much better over the last decade, but even a dilettante like me who’s only been peripherally involved in the industry for a dozen years now can see the difference readily. So you either need a book that you’re going to sell thousands of copies of, or you need an audience that’ll pay upwards of $100 a throw, or both, in order not to lose money. That’s why you see almost no self-published coffee-table books.