The CMS experiment is expected to produce, at design LHC luminosity, few Peta Bytes of data per year and distribute them globally through the LHC Computing Grid.Within the CMS Tier-based infrastructure, some of the user tasks can be optimized using local resources hosted in a Tier-3.In this talk we present CMS Tier-3 Perugia experience in allowing users to perform physics analyses. We describe the local data center organization and its distinctive features, we show the site's bottlenecks in supporting CMS local activities and the workarounds implemented.Finally, we discuss current developments and future ideas for the local workflows optimization, taking into account Tier-3 limits and needs.