Video editing methods and systems enable an editor to edit a video project for which source media assets are located at a media storage server located remotely from the editor with substantially the same fidelity and editing feature set that would be available if the source media assets and editor were co-located. A video editing client used by the editor maintains a persistent cache of proxy media with the layers of the video project stored independently, facilitating editing with combinations locally originated assets and remote assets. The client requests frames not already cached from the remote server via a low bandwidth network. Unless a frame is purged from the cache, no frame is requested from the server more than once. A multi-level priority prefetching scheme, including sequence-based prefetching, populates the cache with frames likely to be requested during editing.