Schelling segregation is a well-established model used to investigate the dynamics of segregation in agent-based models. Since we consider segregation to be key for the development of political polarisation, we are interested in what insights it could give for this problem. We tested basic questions of segregation on an agent-based social network model where agents' connections were not restricted by their spatial position, and made the network graph much denser than previous tests of Schelling segregation in social networks. We found that a dense social network does not become as strongly segregated as a sparse network, and that agents' numbers of same-group neighbours do not greatly exceed their desired numbers (i.e. they do not end up more segregated than they desire to be). Furthermore, we found that the network was very difficult to polarise when one group was somewhat smaller than the other, and that the network became unstable when one group was extremely small; both phenomena may help explain the complexity of real-world polarisation dynamics, such as unique risks faced by very small group sin a society. Finally we tested Fossett's (2006) "paradox of weak minority preferences", a well-established result in grid- and map-based models which shows that an increase in the minority group's desire for same-group neighbours can create more segregation than a similar increase for the majority group. In a densely connected social network, we find that the evidence for this effect is mixed.