This test measure’s the time equilibrium of a connection as bandwidth consumption rises. What does this mean? If you ask how long does it take to drive to the airport, the reply will likely be ‘it depends on when you go’. Understanding time clearly defines the user experience, the conditional reply is needed because time changes based on traffic volume. A reply of the road can manage 75 cars per minute is of no help. Essentially a bandwidth efficiency test measures a connection’s ability to deliver data on time, i.e. the user experience. A prefect result is around 94% at full capacity, a reasonable result starts at 75%.

Nearly all so-called "speed" tests report that ten cars doing 1mph is the same result as one car doing 10mph. Is this true?.... Consider the user experience driving ten miles to catch a flight that departs in 2 hours, the 10mph car will arrive in 1 hour and catch the flight, but the 1mph cars will arrive in 10 hours and miss the flight. In short, any test that reports ‘10 cars arriving in 10 hours’ as ‘1 car per hour’ is a bad test. The efficiency of the 10 cars is 10%, the efficiency 10mph car is 100%. The user experience is dictated by time not bits per second.

Absolutely it is. A low efficiency results means a lot of data is delayed. A low efficiency connection is like a road that is always congested in a rush hour, arriving on time is highly unlikely. A ‘cars per hour’ rating is meaningless to the user experience, whereas arriving on time (efficiency) is everything to the user experience.

If a connection is heavily provisioned the efficiency can fluctuate a huge amount. The best human example is to consider a road that has a lot of traffic signals. Obviously the job of a traffic signal is to provision the flow of cars by turning red then green. If you compare any two journeys where one journey had more red than green then the 'more green journey' will likely exceed the throughput of the journey that had more red.

Essentially the efficiency has increased.

Some providers use provisioning to ramp the connection, either ramp up or ramp down (most common is ramp down). This provisioning process allows you more throughput initially and then it slows. Ramping down gives the user a false sense of a good service when doing lots of small browsing interactions (most common) versus a big download. Ramping up is the opposite, starts slow then goes faster, this is better because small browsing transactions that are slower don't really get noticed. But when you do a big download transaction ramping up will save you lots of time.