quite a few people have seen that one, where fetch grabs a file, and the main engine starts building while the file is not there yet.
What actually happened is that there was some misguided caching optimization with respect to fetch objects, and the object was marked as okay as soon as we asserted the checksum was alright... and the main engine would fire up the build. BUT fetch actually stages the fetch into a file.part temporary file, and moves it to the actual file name as a last step (*after* the checksum was verified), so you ended up with the object saying "okay, I'm cheksummed (the temp file is okay), the build engine starting the build, re-checksumming the same file (hey, it's not in the cache) and either failing (okay the file has not been moved yet) or checksumming it for nothing (the rename and cache move happened too late)... The diagnostic specifics took longer than the actual fix... *IF* you still see fetch trouble in dpb, then that would *definitely* be NFS's fault. :)