There isn't currently a method to import documents from a stream rather than a file.
If you could parse the data you need from a GenBank format string you could create a SequenceDocument and add that to your Geneious folder using WritableDatabaseService.addDocumentCopy(), but I'm guessing that the conversion of, e.g., a BioJava RichSequence to a SequenceDocument is more trouble than it's worth.
Have you really found that writing Genbank records to temp files is a huge bottleneck? How many sequences are you moving at once? In practice we find that disk IO isn't so limiting as long as it's done in a background thread.