If you mine uranium ore and smelt it into uranium metal, you end up with uranium which is mostly U-238 and about 0.7% U-235. This is the natural distribution of isotopes; there are a few others besides those two but they're the majority. (U-233, for example, is some piddlin' little fraction that's so small it's not worth mentioning here.)
It's possible to "enrich" uranium--to increase the fraction of U-235--by running it through various machines which separate the two isotopes, and given enough time and machinery you can take a lump of uranium and make two lumps, one of pure U-238 and one of pure U-235. "Weapons grade" U-235 is about 90% pure.
U-235 is the isotope we find most useful for various kinds of nuclear reactions because of its neutron capture cross-section and other physical characteristics. While you can build a nuclear reactor which uses natural uranium, it's more practical (given that we have the capability to enrich uranium) to enrich it a bit.
Most reactors use fuel that's been enriched to 3% U-235. This isn't nearly enough to make a nuclear weapon, and in fact the unused fuel is so safe the only shielding you need to protect yourself from it is a paper bag. (Not even that, actually--the fuel pellets are encased in zinc cladding, and these little zinc cans are themselves encased in rods. An unused fuel element is no more dangerous than a curtain rod.)
After it's spent time in a reactor, though, there are all kinds of very active isotopes with short half-lives and dangerous emissions (neutrons, betas, gammas, etc) which means you do not want to get cozy with used fuel elements. Here's where the confusion comes in: while it's true that the used fuel element has less U-235 in it than a fresh one, that's not why we stop using them.
We stop using fuel elements because of the buildup of fission products. E=mc2 is what provides the energy--as fission occurs, energy is released--but the energy we extract from nuclear fuel is the barest fraction of a whisper of the total energy available. When any atom splits, there's a tiny bit of energy released but most of the atom's mass remains behind, only in smaller chunks. This is why they call it "fission": the atom's nucleus splits.
It usually does not split into equal chunks, and there's no way to predict what elements a particular atom will split into. It's all random. The important thing, though, is that it's not uranium.
That's important because of that characteristic I mentioned above, "neutron capture cross-section". The easiest way to imagine that is to think of a big brick wall with a window set in it, high up, into which you are trying to throw a ball. If you get the ball through the window, you win a cookie and get another ball to try again.
The problem is, the window is an inch wide, and the ball is a softball.
So you move to a different wall. This wall has a huge French Door set in its side; you can easily throw the ball through the window--but because you moved to the easy wall, you don't get a cookie or another ball.
There's another wall nearby, just behind the easy wall, which has a window about a foot on a side. You would have gotten a cookie and another ball had your ball gone through that one's window, but the easy wall was in the way.
The ball is a free neutron; the wall is an atomic nucleus; and the window is its neutron capture cross-section. The cookie is useful energy of one form or another (up to and including another free neutron). The hard wall is an atom of U-238; the easy wall is a "poison", and the wall behind the easy wall is a U-235 atom.
We can make the ball appear to change size by making it pass through a medium we call the "moderator". It doesn't actually change the ball's size; it changes its speed--but this is a weird quantum ball, and the size of the window is dependent on the speed of the ball. (For the purpose of this analogy we assume you can only throw at one speed, but you get the idea by now, I'm sure.) The slower the ball, the larger the window--up to a point--and passing the ball through the moderator enlarges the windows of all the walls, but it's most noticeable with the U-238 wall. Too slow, of course, and you can't get the ball high enough to go into the window; so you need to change its speed so that it's got just enough to make it into the window.
The function of the moderator is to slow down neutrons, but only enough to get them into the neutron capture cross-section of the reactor fuel. Neutrons come out at a variety of speeds, and it's the slower ones that we find useful--too fast and they just blast through a nucleus without slowing down or affecting anything, and you want the nucleus to capture the neutron.
But you also want the nucleus to give off a neutron or two when it splits, because that's how you sustain a nuclear chain reaction. If the neutron vanishes into a nucleus and nothing comes out, the chain reaction dies out.
This is what happens with used nuclear fuel: stuff builds up in the fuel elements that soaks up neutrons and doesn't give any back, nor does it give off any useful energy. This is why they're called "poisons": they poison the nuclear chain reaction.
Fortunately, all this stuff is not uranium, so it can be chemically separated from the fuel. It's easy-peasy to make chemically pure uranium. ("Isotopically pure" is hard, but we don't need that.)
Once you've finished with that process, you have a small pile of stuff you really don't want to stand next to, but your fuel is ready to go back into the reactor.
The thing is: there is less U-235 in this fuel than there was before you used it, but it's still useful.
"Why enrich it at all, then? Or why enrich it to 3% if less will work?"
A higher concentration of U-235 is useful because it's so good at soaking up neutrons, splitting, and releasing new neutrons. But because most of the fuel (97% of it) is U-238, that's where most of the fission occurs and that's where most of the energy is coming from. The U-235 serves primarily as a jump-starter for the reactor. Its capture cross-section is pretty large, it generally gives off a neutron or two when it fissions, and it's pretty abundant in nature. (Compared to other isotopes? You bet.) You can make a reactor that uses natural uranium--the first one in history, Fermi's "atomic pile" under the bleachers at University of Chicago, was made with unenriched uranium--but there tend to be engineering considerations that make them impractical.
In practice, when reprocessing used fuel, you need to add enough uranium to the fuel elements to make up for what you lost--energy and fission products alike. And that's it. You want to try to keep it around 3% U-235, but that's not as hard as it sounds.
You certainly do not need to take--say--97 pounds of used fuel and add another three pounds of fresh U-235 to it to get back to 3% enriched fuel.
* * *
Opinion: it seems to me that the proportion of isotopes would remain approximately the same, in fact. My example above exaggerates the difference in cross section between U-238 and U-235; they're actually not all that far apart. The distribution of U-235 in enriched uranium is random, and it's only 3% of the total; the speed and direction of emitted neutrons is also pretty random, though speed is confined to a certain range.
A neutron has no choice about what it runs into, and U-238 will fission if it captures a neutron that smacks into it. It therefore seems that the general proportion of isotopes in the fuel will remain approximately the same, and that the total percentage of uranium will simply decrease over time.
* * *
The important thing to remember is that used nuclear fuel is not useless. If you're allowed to remove the poisons from it, you can stick it back in the reactor and use it again.
Right now, we use nuclear fuel the way a rich man with no sense uses cars: "Oh, my, this Ferrari is dirty. Time to buy a new one!" The car only needs a wash and wax, but instead of doing that, the idiot just lets the slightly used cars pile up in his garage.
Then he complains that there are too many used cars around, and that we have to stop making cars because of the problems we're having with storing the used ones.