Yes, accurate measurement time stamps could always be helpful: Velocity calculation traps
Also, depending on the dominating measurement error source averaging 20 consecutive samples may or may not be the best way to get accurate result.
If the errors have normal Gaussian distribution (like thermal noise) then averaging is effective at reducing the mean error.
However, if the source of errors is intermittent, like mechanical vibration that will affect 4-5 of the 300 measurements, but make them 2x times the expected value, then averaging will not eliminate the error but hide those erroneous measurements by diluting it among otherwise good sample. The proper way to handle such errors is to filter them out, based on expected value +/- reasonable range and only average values that make sense physically - i.e. you don’t expect full robot to accelerate twice as fast for 10ms then slowdown to expected acceleration in the next 10ms. That would be an indication of the vibration affecting separate small components.
It would be great if we could get raw measurements from the 3-wire legacy ports. And, if it is not possible due to the packet size, the next best thing would be to have an option to either return a single timestamped raw measurement or an average of several raw measurements along with the standard deviation. This way higher level algorithms could throw it out if they suspect presence of the larger than expected non-gaussian noise.