I found an oldish paper on this topic (from 1994). Here's a summary:

Determination of the optimal aligned spacing between the Shine-Dalgarno sequence and the translation initiation codon of Escherichia coli mRNAs. by Chen, Bjerknes, Kumar, & Jay. Nucleic Acids Research. (1994)

Experiment

The authors constructed a series of synthetic RBS regions that varied the length separating a synthetic 5-nt Shine-Dalgarno sequence from the start codon. The regions varied in size between 2 to 17 nt. They assayed the activity of a downstream enzyme, chloramphenicol acetyltransferase.

Conclusion

The authors concluded that the optimal spacing between their consensus 5-nt Shine-Dalgarno sequence (5'-GAGGT-3') and the start site was 5 nt. They also referenced other several other authors who had conlcuded 5 nt was optimal by looking at natural mRNAs from E.coli.

Details

• the RBS is considered to be large, extending 20bp on either side of a core Shine-Dalgarno (SD) sequence. These days, I often hear of the RBS spoken of in sizes that are equivalent to the SD. So in the parlance of the paper, you question is rephrased as "how does distance of the SD from the start codon effect translation?"

• the canonical SD sequence referenced in the paper is 5'-UAAGGAGGU-3'. It is 9 nucleotides long. Distances between the SD and the start codon are defined as the number of nucleotides separate the 3' Uracil of the SD from the Adenine of the start AUG.

• Example: the distance is 5 nt in the following mRNA

          5'.....UAAGGAGGUnnnnnAUG......3'
  

• if the SD is not a complete 9 nt long, the distance is between the position of where the canonical Uracil would occur. In the following example, the distance is still calculated as 5 nt:

          5'.....UAAGGAnnnnnnnnAUG......3'
  

• the average distance between the SD sequence and the start codon varies considerably and on average is 7 nt. Other investigators have found "optimal" spacing (circa 1994) ranged from 5 to 13 nt.

• the SD site complements with region on the 16s rRNA. The start codon complements to the anticodon of fMet-tRNA loaded into the ribosomal P-site. So there are two distinct sites on the ribosome that contact the mRNA during translation initiation.

Also good to read

• Designing Genes for Successful Protein Expression by Welch et al. Methods in Enzymology, Vol 498. (2011) (pdf)
• Automated design of synthetic ribosome binding sites to control protein expression by Salis, Mirsky & Voigt. Nature Biotechnology. (2009). (pdf)

Note

I enjoyed researching this question because I found my own knowledge to be lacking solid empirical details. I do not have any direct experience besides this little lit review with this topic.

I found an oldish paper on this topic (from 1994). Here's a summary:

Determination of the optimal aligned spacing between the Shine-Dalgarno sequence and the translation initiation codon of Escherichia coli mRNAs. by Chen, Bjerknes, Kumar, & Jay. Nucleic Acids Research. (1994)

Experiment

The authors constructed a series of synthetic RBS regions that varied the length separating a synthetic 5-nt Shine-Dalgarno sequence from the start codon. The regions varied in size between 2 to 17 nt. They assayed the activity of a downstream enzyme, chloramphenicol acetyltransferase.

Conclusion

The authors concluded that the optimal spacing between a consensus 5-nt Shine-Dalgarno sequence (5'-GAGGT-3') and the start site was 5 nt. Note: this synthetic SD was made of the last 5 nt of a 9 nt SD consensus sequence. They also tested a synthetic SD made from the first 5 nt (5'-TAAGG-3')of the consensus SD. In this case they found the optimal distance was 9 nt.

So the optimal distance depends on where your desired SD aligns with the consensus SD sequence, which optimally is 5 nt from the start. Read on for more details.

Details

• the RBS is considered to be large, extending 20bp on either side of a core Shine-Dalgarno (SD) sequence. These days, I often hear of the RBS spoken of in sizes that are equivalent to the SD. So in the parlance of the paper, you question is rephrased as "how does distance of the SD from the start codon effect translation?"

• the canonical SD sequence referenced in the paper is 5'-UAAGGAGGU-3'. It is 9 nucleotides long. Distances between the SD and the start codon are defined as the number of nucleotides separate the 3' Uracil of the SD from the Adenine of the start AUG.

• Example: the distance is 5 nt in the following mRNA

          5'.....UAAGGAGGUnnnnnAUG......3'
  

• if the SD is not a complete 9 nt long, the distance is between the position of where the canonical Uracil would occur. In the following example, the distance is still calculated as 5 nt:

          5'.....UAAGGAnnnnnnnnAUG......3'
  

• the average distance between the SD sequence and the start codon varies considerably and on average is 7 nt. Other investigators have found "optimal" spacing (circa 1994) ranged from 5 to 13 nt.

• the SD site complements with region on the 16s rRNA. The start codon complements to the anticodon of fMet-tRNA loaded into the ribosomal P-site. So there are two distinct sites on the ribosome that contact the mRNA during translation initiation.

Also good to read

• Designing Genes for Successful Protein Expression by Welch et al. Methods in Enzymology, Vol 498. (2011) (pdf)
• Automated design of synthetic ribosome binding sites to control protein expression by Salis, Mirsky & Voigt. Nature Biotechnology. (2009). (pdf)

Note

I enjoyed researching this question because I found my own knowledge to be lacking solid empirical details. I do not have any direct experience besides this little lit review with this topic.