I've seen this a bunch in Patchwork but never known why -- now I know!

During syncing most flumeviews will save their progress to disk, so if you restart the server while indexing it will start up where it left off. This doesn't seem to be the case with flumeview-reduce, which doesn't seem to save its progress until it's totally up-to-date.