I think it really helps if you have a physics background to help you understand analogies that are totally valid but not exactly like the things they represent. It's not the kind of thing high school grads or people with liberal arts degrees can grasp right away.
Physicists use exaggerated examples to understand and explain how things work. I'm sure engineers do it, too. For example, if you want to understand how a system behaves when one item's mass is changed, but you need a quick answer without much math, about principle, not exact numbers, you may pretend the mass is infinite or zero, even though these examples could never happen, and this answers your question instantly. That would be called using "limiting cases." I have forgotten most of the physics I learned before leaving grad school many years ago, but I recall how exaggeration is used to make systems easily understandable, and I have a pretty good understanding of buoyancy, which is very simple as physics go. It's the kind of thing doctors learn in their brief and very limited exposure to basic physics, which is all most of them learn. I used to teach them.
The oil example I gave is exaggerated in order to help people who don't know much about science understand it, but it is valid for illustrating a principle that indisputably exists. It is easier for most people to understand an extreme example of oil and water than a fluid which has a smooth density gradient, like stout that has settled.
In the case of beer with heavy ingredients concentrated toward the bottom, and the case of oil and water, you have a liquid sample which is denser at the bottom than the top, so a hydrometer will behave similarly, not identically, in both cases. The principle is the same. Same kind of error.
The bulb of the hydrometer displaces more fluid per unit of length than the skinny part, and therefore more mass, even when a fluid is homogeneous. When a fluid is heavier toward the bottom than the top, the bulb's contribution to the reading is even greater than it is in a homogeneous fluid.
When heavy ingredients move down in a beer, you get a heavier fluid toward the bottom, where the bulb is. This pushes the hydrometer up and gives you a reading that is not representative of the fluid as a whole. To get an average reading, you have to shake the fluid so the density is uniform. This would be true even with a refractometer, because heavier wort has a higher index of refraction.
If you don't like oil and water, think of a HYPOTHETICAL fluid with a density gradient that is extreme. Say the density increases with the cube of the distance from the meniscus. Up high, it may be like water. Down deep, it may be more like mercury, which is denser than lead. The heavy fluid at the bottom will exert a disproportionate amount of upward force on the hydrometer because it's down where the bulb is. That gives a high reading that isn't the average density of the fluid.
If it works in an imaginary extreme case, it will work in ordinary wort.
The business with poodles is not helpful because poodles are large and discrete. Beer is full of tiny suspended items and things that are sort of like, but not really, liquids. Think of the milk example Larry Shepley mentioned. Milk is full of tiny globules of fat until it separates. These globules aren't actually dissolved, but they are so small, the shaken milk acts like a homogeneous solution (hence "homogenized") until it separates. At first, you have a uniform sample, and a hydrometer works perfectly. Then you have a sample with a density gradient that changes with time. Then you end up with two liquids with a defined border. During the last two stages, a hydrometer will not work, and by "work," before someone says something that isn't right, I mean work to give you the average density of the entire sample.
If you don't believe I understand how this works, believe Larry did. He was a pretty smart guy. He was a protege of John Wheeler, and he went from a master's to a Ph.D. in two years. Easy in history, literature, and sociology. Not easy in physics. It's actually more impressive than it sounds, because Princeton used to give all grad students master's degrees as soon as they were accepted, so he went from undergrad to Ph.D. in two years.
Poodles float, by the way, so not really useful at all.
Here's something interesting, at least to me. Remember the scene in "Aliens" where Ripley dropped herself in molten lead and sank in? Couldn't happen in real life. She would have rolled around on the hot surface in agony because she couldn't sink more than a couple of centimeters, and she would have broken a lot of bones on impact because the mass of the lead would have resisted accelerating away from her body. She was not dense enough to sink.