Why not convert to integer (saving residuals), computing mean of integers via long division and adding means of residuals (arbitrary precision sum)?