Let $X$ be a (separated, and with whatever other tameness conditions are appropriate) scheme over the integers $\mathbb{Z}$. (If you don't like schemes much, imagine that $X$ is described by algebraic equations over the integers, so that it is possible to make the algebraic variety $X(k)$ for any field $k$.) I am interested in the case in which $X(k)$ is "the same" for all fields $k$ so that it is possible to discuss its properties over the "field with one element $\mathbb{F}_1$". (Actually I think that $\mathbb{F}_1$ is a scam to some extent, but I wanted to make the question catchy.) What I mean is examples such as when $X$ is a Grassmannian. In this case the Poincaré-Hilbert polynomial of the topological cohomology $H^*(X(\mathbb{C}),\mathbb{Q})$ is the same up to the change of variables $q^2 \mapsto q$ as that of the cohomology $H^*(X(\mathbb{R}),\mathbb{Z}/2)$, and that's the same as the polynomial expression of the number of $\mathbb{F}_q$-rational points of $X(\mathbb{F}_q)$, and probably a lot of other things are the same. They're all the same because the Schubert cell decomposition is perfect in the sense that it has a vanishing differential over either $\mathbb{C}$ or $\mathbb{R}$, and it has equally nice properties over any other field.

Is there a name for a scheme with this type of good behavior? The question is not entirely rigorous, except in the primitive form of reducing only to the fields listed above. Even so, most schemes clearly don't share these stability properties possessed by Grassmannians. For instance, the picture fails badly for an elliptic curve.

I got two good answers to the question (and a third one that was not bad). In the interest of wrapping things up, I would like to accept one of them. The answers are basically tied in mathematical value, but I particularly like Jordan Ellenberg's prose. The picture from both answers taken together is that people have worked both on upper bounds --- all schemes that could possibly qualify for the question --- and lower bounds --- schemes that by design must be allowed as examples. The former are called "polynomial count" varieties; the latter are sometimes offered as varieties defined over $\mathbb{F}_1$. Pure Tate cohomology is another type of upper bound. It seems that the discrepancies between the upper and lower bounds are not well understood.