In recent years, massive datasets such as those from the Digital Michaelango Project have inspired the development of out-of-core (OoC) algorithms that process surfaces in a spatially coherent order. This reformulation of surface processing yields mesh data structures which are more cache-friendly than prior representations. Although OoC techniques are intended to mask high latency disk access, the same principles can benefit processing in other levels of the cache hierarchy. This survey gives a background on traditional geometry processing, examines the development of OoC mesh processing algorithms, and briefly considers the potential of OoC-like data structures for parallel- and stream-processing.