also because we know the files that will be written to in advance in most cases (

output_directories

notwithstanding) i personally believe we should be able to get ridiculously faster-than-native speed on writes for processes executed within the VFS