Luminous Red Galaxies (LRG) from the Sloan Digital Sky Survey are considered among the best understood samples of galaxies, and they are employed in a broad range of cosmological studies. Because they form a relatively homogeneous population, with high stellar masses and red colors, they are expected to occupy halos in a relatively simple way. In this paper, we study how LRGs occupy massive halos via direct counts in clusters and we reveal several unexpected trends suggesting that the connection between LRGs and dark matter halos may not be straightforward. Using the redMaPPer cluster catalog, we derive the central occupation of LRGs as a function richness, Ncen({lambda}). Assuming no correlation between cluster mass and central galaxy luminosity at fixed richness, we show that clusters contain a significantly lower fraction of central LRGs than predicted from the two-point correlation function. At halo masses of 10^14.5 Msun, we find Ncen=0.73, compared to Ncen of 0.89 from correlation studies. Our central occupation function for LRGs converges to 0.95 at large halo masses. A strong anti-correlation between central luminosity and cluster mass at fixed richness is required to reconcile our results with those based on clustering studies. We also derive P_BNC, the probability that the brightest cluster member is not the central galaxy. We find P_BNC ~ 20-30% which is a factor of ~2 lower than the value found by Skibba et al. 2011. Finally, we study the radial offsets of bright non-central LRGs from cluster centers and show that bright non-central LRGs follow a different radial distribution compared to red cluster members, which follow a Navarro-Frank-White profile. This work demonstrates that even the most massive clusters do not always have an LRG at the center, and that the brightest galaxy in a cluster is not always the central galaxy.