The short answer is no, slight violations of these ideas don’t tend to cause the sky to fall. Indeed, the list of examples given by nikatwork was unfortunate, because some of them are well established with a decent evidence base, but a few are basically just nonsense that has unfortunately somehow become dogma.
For example, a few studies have been performed under controlled conditions to try to determine how typographic details like line length affect measurable performance levels such as reading speed and retention. There are many rules of thumb floating around — “2.5 alphabets”, “8–10 words”, and the like — and it actually does turn out that most of them are reasonable. It also turns out to be unwise to consider line length in isolation, that a wider range of lengths can be used before serious degradation is widely observed than some of the old folklore would suggest, and that having lines too short can also cause serious degradation in reading performance, among other interesting results.
Another good example is the consistency of user interface elements. Numerous usability tests, from formal studies to simple A/B tests on web sites, support the theory that violating a user’s expectations tends to have negative effects, sometimes severe ones. Of course you always have to be careful not to assume too much about what your users’ expectations might be, but things like using green for cancel and red for confirm almost invariably exhibit poor performance with audiences who are used to the opposite convention.
On the other hand, if someone tells you the Golden Ratio is important for keeping some sort of page area proportionate, it’s a good bet that this person should be hunted down and locked away by the maths police. Using precise ratios really does matter in some contexts: if you look at the ratio between the width and height of a piece of A4 paper, it is aiming for exactly √2, which is why two A4 sheets fit neatly into one A3 sheet and so on up and down the scale. However, someone who is advocating use of the Golden Ratio for width/height proportions usually just likes the look of dimensions that are proportioned roughly 5:3, and probably has no hard data to support choosing that over 3:2, 16:9, 1:1, or any other ratio that fits with the rest of the design.
Likewise, if someone advocates picking a colour scheme based on where certain hues fall on a wheel, consider that hue represents only one dimension. The most common colour models for human vision use three, and in practice if you fix the others (typically saturation and lightness) and just choose evenly spaced and/or opposing hues around a colour wheel, the results are rarely good. That means much repeated ideas like choosing analogous or complementary colours are, at best, only part of the story. If you’re using a typical colour wheel for computer graphics that has red, green and blue at 120 degree separations and then cyan, magenta and yellow on the 60s, those ideas are even less helpful, because it turns out that by the time the human eye has detected the light and the human brain has interpreted the resulting signals, those colours aren’t really uniformly spaced around the wheel anyway; see the trichromatic vision and opponent process theories, or the early experimental work of Munsell.
In short, there really is robust evidence that some established design principles are worth following, though rarely do minor variations in design cause major variations in performance. However, there are also some dogmatic assumptions in the design world that get passed on as if they are inalienable truths but in reality have little evidence to support them.