In general, best practice is NOT to pre-translate a file in advance. It is better to use the default behavior of Memsource Editor, which will always pre-translate the next segment you jump to (after confirming the previous segment).
In this case you get the best match available at that moment. If you pre-translate the whole file in Memsource Cloud in advance, it could be that you lose some matches that got created between the moment of batch pre-translation and the actual moment when the translator arrives at that segment (e.g. translated by someone else or even translated by the same translator earlier on in the file).
Sometimes it is of course useful to pre-translate in advance, for instance when you want to work offline or when you want to pre-translate just 101% matches.