Question is rather self-explanatory.
Putting aside other post-transcriptional factors like rate of degradation of transcript, what is the smallest gene ever reported to have successfully been transcribed in E. coli or other similar prokaryotic systems? Why couldn't this gene be any smaller?
Note that in the context of this question, I am disregarding factors like promoters, operators and terminators. Also note that this gene does not necessarily need to code for a protein.
In terms of naturally-occurring "genes" I think that the record is probably held by the attenuator peptides.
In bacteria, the regulatory mechanism known as transcription attenuation involves the ribosomal biosynthesis of very short peptides. In the trp operon, where the phenomenon was first described by Yanofsky's group, the peptide, MKAIFVLKGWWRTS, consists of just 14 amino acids. Other examples, described subsequently, are of comparable size.
Now that I have written this answer, I have reread your question and I see that you are really interested in transcripts. In the case of the attenuators, the peptides are encoded by the 5' end of a much larger transcript, so they probably don't count for you.
Non-coding transcripts can be as small as 10-15 nucleotides. Once the RNA polymerase initiates, the DNA is melted and the transcription bubble forms. This region is about 10 nucleotides long.
But if you are asking the limit of mRNA length, then the answer is different. First of all, to get a protein, the ribosomes have to translate the transcript. To do that, they need some space to land on, and initiate the translation process. We know that ribosomes protect about ~30 nucleotides from ribosome footprinting experiments
Moreover, bacteria wouldn't need to utilize the genomic space and the ribosome machinery to produce small peptides (even if the peptides were useful), since the small peptides are already around as the degradation product of other proteins. Hence, once you add 3 times the shortest produced polyaminoacid length to the minimum length required for the ribosome landing, you get somewhat physical (or let's say physiological) limit of the transcript length.
