The codon translation rate is usually assumed to be proportional to the cellular concentration of the cognate tRNA, but synonymous codons sharing the same cognate tRNA may be translated at rather different rates. To account for the latter observation, we assume that the translation process is optimized in two respects: (i), the codon demand is optimized with respect to the supply of cognate tRNAs (composition of the tRNA pool); and (ii), for synonymous codons sharing the same cognate tRNA, the usage frequency of each codon correlates optimally with the stability of the codon-anticodon complex. These assumptions allow us to compute the relative rate constants of synonymous codons. Highly expressed genes, which produce 80-90% of the protein mass in the E. coli cell, appear to have selected codons which make an optimal use of the tRNA pool. Assuming the optimization criteria were valid, a list of codon translation times (in ms) were derived from available experimental data.