The above answer is correct that FME won't save the current state of a translation if you stop it. Once a translation is started you have to see it through to the finish.
I will file an enhancement request with the developers to see what they think. But I'd also like to discuss or suggest a few other options.
1) Have you had anyone check your workspace to see if it could run in less that that time? I know bulk raster handling can take ages, and sometimes there's no way to avoid a lengthy process, but you can always send your workspace and log file to safe.com/support and ask them to check. Or post it here if you like.
2) My thoughts about improving performance (without having seen your workspace) are that fanouts are expensive in terms of resources, so try to avoid those. So is reading (say) 1,000 raster files together, reprojecting them all, and then writing them back out to 1,000 new files. Maybe you can restructure the translation or reconfigure it in a way that avoids masses of data. For example, process one file at a time using a Master/Slave setup and the FMEWorkspaceRunner transformer. You can even run multiple slave processes simultaneously in such a scenario. See this example to find out how this is done.
3) You want to stop the translation because the machine is needed for something else, so have you considered FME Cloud? That way you have a dedicated machine that is used for FME and you won't need to interrupt it for updates, etc.
I hope something here is of use. Feel free to contact me directly, or post the same as a question on our knowledge centre, where other FME experts can chime in.