The past few years has seen a significant advancement in the predictive capability of constraint-based models of metabolism due to incorporation of thermodynamic constraints. These models allow us to broaden the scope of prediction and analysis to both reaction fluxes and metabolite concentrations. Taken together, these process variables can now provide insight into additional characteristics of metabolism, such as the turnover rate of each metabolite. Furthermore, these in silico investigations can be refined using increasingly available multi-omics data. Using a thermodynamically constrained metabolic model of Escherichia coli with additional constraints from multi-omics measurements in the literature, we found that the metabolite turnover rates were bimodally distributed. We also found a correlation between a metabolite's turnover rate and its connectivity. Furthermore, we investigated the degree to which perturbations in each turnover rate affected the rest of the network. For this, we distinguished between perturbations affecting turnover rate values--but not necessarily their distribution--with those affecting the distribution of turnover rates across the entire network. To account for the variability in fluxes and metabolite concentrations in the above analyses, we formulated a fractional program that could be used to define the maximum and minimum ranges of metabolite turnover rates. Based on the fractional programming literature, we reformulated this problem into a mixed-integer linear program that could be efficiently solved for genome-scale metabolic network models. In addition to offering further insights into network properties, we show how metabolite turnover rates have important implications when designing strains for metabolic engineering applications.