We examine the main drivers of software maintenance effort and cost. We use the 'Distributed Cognition' framework to hypothesize about how 'discovery work' in maintenance is effected by two types of cost drivers: system attributes (size, complexity, age, etc.) and personnel attributes (number of maintainers, location dispersion, etc.). We test our hypotheses using archival data about over 5,000 maintenance projects carried out between 2009 and 2011 on 412 different operational systems in a large financial institution. We find that personnel attributes are significantly more influential than system attributes. In particular, a marginal change in personnel factors is associated with effort growing much faster than cost, indicating an escalating marginal cost of spreading maintenance work across more maintainers and site locations. We also find, counter to expectation, that two system attributes are negatively linked to maintenance effort and cost. Implications of these findings for research and practices are discussed.