Besides platter problems, the #1 reason I've seen for incomplete commit is what I can only describe as "confusion" with the Thick Client's workstation options. It makes sense what's happening, but it's still annoying if your environment has higher than normal security.
What happens is, when you set the Config --> Global Client or Client --> Workstation Options to use Core workflow (which is the default), any and all Thick Client instances including services will use that setting to do work. Well, the problem with Thick Client's Core workflow is that it really does like to time you out rather frequently. If this has happened the service will fail to do its job if the document type is configured to go into workflow. Additionally, there's no way to tell that it's what's happened from error logs. Restarting the services or running the jobs manually after restarting a client then works because it gets a new session.
This has forced me to make sure that my processing server with all of the services remains locked at Classic workflow to ensure that I don't encounter this problem, at least until such a time as when I can figure out how to get the Thick Client not to time out of Core workflow.
