In this paper, we consider the problem of efficiently representing a set $S$ of $n$ items out of a universe $U=\{0,...,u-1\}$ while supporting a number of operations on it. Let $G=g_1...g_n$ be the gap stream associated with $S$, $gap$ its bit-size when encoded with \emph{gap-encoding}, and $H_0(G)$ its empirical zero-order entropy. We prove that (1) $nH_0(G)\in o(gap)$ if $G$ is highly compressible, and (2) $nH_0(G) \leq n\log(u/n) + n \leq uH_0(S)$. Let $d$ be the number of \emph{distinct} gap lengths between elements in $S$. We firstly propose a new space-efficient zero-order compressed representation of $S$ taking $n(H_0(G)+1)+\mathcal O(d\log u)$ bits of space. Then, we describe a fully-indexable dictionary that supports \emph{rank} and \emph{select} queries in $\mathcal O(\log(u/n)+\log\log u)$ time while requiring asymptotically the same space as the proposed compressed representation of $S$.