Do the performance gains depend on how many digits before or after the decimal point there are?
No, this encoding uses a bitwise encoding for each float, so they are fixed at 10 characters each and encoded with constant speed (except possibly for subnormal numbers, which I've heard are slower in some environments).