Are there arbitrarily sparse "lattices" in negatively curved symmetric spaces?

The examples are arithmetic groups, constructed in general by Borel and Harish-Chandra. See also Dave Witte Morris' preliminary book. However, examples in hyperbolic and complex hyperbolic spaces probably go back further to the study of quadratic forms.

For hyperbolic lattices, one can take a quadratic form over a quadratic number field (such as $\mathbb{Q}(\sqrt{2})$), which is Lorentzian at one place, and definite at the other places (such as $x_1^2+\cdots +x_n^2-\sqrt{2} x_{n+1}^2$), and take the group of matrices $\Gamma$ in $GL(n+1,\mathbb{Z}[\sqrt{2}])$ which preserve this quadratic form. Then Mahler's compactness theorem (cf. Witte Morris) implies that the quotient $\mathbb{H}^n/\Gamma$ is compact. Then by Selberg's Lemma and residual finiteness (as Greg points out, Malcev's Theorem), you may find a torsion-free subgroup of finite-index with as large injectivity radius as you like.

For hyperbolic and complex-hyperbolic spaces, there are other examples which don't come from the arithmetic construction (in fact, most hyperbolic surfaces and 3-manifolds are not arithmetic). These are attributable to Gromov and Piatetski-Shapiro in the hyperbolic case in all dimensions, and there are finitely many examples in the complex hyperbolic case (at least for $\mathbb{C}$-dim >1) going back to Deligne and Mostow (see also Thurston's paper). However, Gromov and Schoen have shown that quaternionic and Cayley-hyperbolic lattices are all arithmetic, so the Borel Harish-Chandra construction is complete in these cases.

Addendum: Related to Protsak's comment, there is a simple example of a negatively curved homogeneous space which has no lattice action. This is Thurston's "9th geometry", which is excluded as a geometry precisely for this reason. One can take the double warped product metric $$ dr^2 + e^{2a r} dx^2 + e^{2b r} dy^2, $$ for $a,b >0$. When $a=b$, this gives hyperbolic space. But when $a\neq b$, the sectional curvatures are $-a^2, -b^2, -ab$. This has a solvable transitive group of isometries, so is homogeneous. But using the solvability, one may see that there is no cocompact action. (Remark: when $a=-b$, this gives sol geometry).

Let me give an explicit example for the Poincare disk in dimension 2. In the disk model, just take the regular $4g$-gon centered on the origin. Identifying opposite sides, we get a compact surface of genus g. The polygon is a fundamental domain for the group $\Gamma$ generated by $b_1...b_{2g-1}$ with

$b_k=\pmatrix{\cosh(a) & e^{ik\pi\over 2g}\sinh(a)\cr e^{-ik\pi\over 2g}\sinh(a)& \cosh(a)}$

where a is defined by $\cosh(a)=1/\tan(\pi/4g)$. The value of a is in fact equal to the radius of the maximal disk contained in the fundamental domain. This is also the distance from any side of the 4g-gon to the origin. It goes to infinity with g, so we are done.

The Gauss-Bonnet formula tells us that surfaces of a given genus have a bounded volume, hence cannot contains arbitrarily large embedded balls. Hope that helps.