Theoretical considerations of Bell-inequality experiments usually assume identically prepared and independent pairs of particles. Here we consider pairs that exhibit both intra- and inter-pair entanglement. The pairs are taken from a large many-body system where all the pairs are generally entangled with each other. Using an explicit example of single mode entanglement and ancillary BEC, we show that the Bell-inequality violation in such systems can display statistical properties that are remarkably different from those obtained using identically prepared, independent pairs. In particular, one can have probabilistic violation of Bell's inequalities in which a finite fraction of all the runs result in violation, even though there could be no violation when averaging over all the runs. Whether or not a particular run of results will end up being local realistically explainable is "decided"by a sequence of quantum (random) outcomes.