The batch effect would indeed be a major problem. You can remove batch effects when you have sample replicates spread across the different batches (and account for it in the model matrix in limma) but if that's not the case, it may be difficult or impossible since the batch effects could completely confound your biological difference. It sounds like you may have too many sources of non-biological variation like samples coming from different labs, different tissues, different times, chips and scanners etc., even if the data has all been derived using the same platform.