While it is well established that highly expressed genes in bacteria exhibit stronger codon optimization than lowly expressed ones, whether codon-anticodon adaptation shows fine-scale differentiation among highly expressed genes themselves remains unexplored. Ribosomal proteins, which are expressed in stoichiometric amounts and often co-transcribed in polycistronic operons, provide an ideal system for testing such differential selection. Here, I demonstrate that in Escherichia coli, Bacillus subtilis, and Vibrio natriegens, codon usage is more optimized in long ribosomal protein genes compared to short ones. This pattern persists even among genes within individual operons such as S10, spc, and α operons. A ribosome in E. coli or B. subtilis needs four copies of L7/L12 (encoded by rplL) but only one copy each of the other ribosomal proteins. This high demand for L7/L12 leads to my prediction that the rplL gene should be translated more actively than its operonic partner, rplJ, encoding L10. This prediction is also strongly supported by empirical evidence from representative bacterial species. Actively translated mRNAs are protected from endonucleolytic cleavage and degradation. If rplL mRNA is more actively translated than rplJ mRNA, then rplL mRNA would be degraded less and become more abundant than rplJ mRNA, which is true. These results demonstrate that translation optimization reflects functional stoichiometry and protein length constraints. This is the first demonstration of natural selection operating predictably and precisely among ribosomal protein genes in the same operon, fine-tuning translational output to achieve efficient ribosomal assembly.