Total power consumption is about 7:1 when comparing states of moderate physical labor or 9:1 when comparing idle states.
A Boston Dynamics Atlas robot is a humanoid robot with a human like body mass of 190lb (85kg). It reportedly can run for about 1 hour using its 3.7kW/hr battery doing normal physical exertion including walking, standing, use of tools, and other movements or it can idle for about 4 hours.
According to this source, a human in this same weight range would consume about 450 kcal per hour performing similar activities which translates to 0.52 kW/hr or about 85 kcal/hr (0.099 kW/hr) when awake and doing basic sensory processing like watching a TV show or reading. This means that if you were to replace all of the systems on a current Atlas android with something as efficient as biology, you could get approximately 7 hours of work or 37 hours of idle time out of the battery packs we already use.
So no, you will not get decades of battery life (you definitely had a math error here), but you are correct that biology is significantly more energy efficient than modern robotics.
Why muscles are so much more efficient than cybernetics
Even though the electrical motors that we use to move a robot tend to be over 75% efficient in terms of converting electrical energy into kinetic energy, there are a lot of sources of inefficiency outside of the actual motors.
When it comes to making a motor that turns electricity into rotational energy, like turning the wheels on a car, mechanical engineering is pretty reliable, but moving something android shaped means that you need to perform a lot of contractions and rotations in ways that don't leave room for motors at the point of articulation; so, to engineer an android using electric motors, it means that we need to use a lot of hydraulics, belts, gear systems, axils, and other intermediate systems to translate the energy from the motors into the kinds of articulation that we need.
However, when we compare this to a muscle, the entire mechanical system between 2 points IS the motor; so, there is no wasted mass and resistance from intermediate mechanical systems.
But Why Do Humans Idle Better Too?
A human sitting around is doing all sorts of complex metabolic stuff, so the idea that robots, without metabolisms, consume so much more power than we do when just sitting around sounds counter-intuitive, but it is true.
For starters, just standing or sitting will still consume some energy. It takes energy hold an electric motor in place against the resistance of gravity. It takes energy to perform the micro adjustments to rebalance your body when trying to stay upright against changes in the environment. It takes energy for your robot to run all of the cameras, microphones, gyros, pressure sensors, and switches required for it to understand its environment and position in space.
There is also all of the power lost to induction currents. Electricity does not just stop flowing because you turn off a switch, the switch creates an air gap that makes the electricity flow much more slowly. Even if you turn an average robot completely off, if you do not unplug the battery, it's charge will run out within a few weeks as electricity jumps the gap creating by the power switch.
There is also the problem with robots that they don't think nearly as efficiently as we do. The human brain a true neural network, but modern robots use linear digital networks to simulate neural networks which is a grossly inefficient way of doing it. So, the power it takes a robot to process it's inputs is higher than that of a human brain. So, if you want a robot to be physically in standby, but still processing its environment, it will use a lot more thinking power than a human does.
How to make an Android as efficient as a human
To make an android able to do human things as efficiently as a human, you first need a fibrous material that contracts when electrically stimulated. There has been a significant amount of research in recent years studying Electroactive polymers (EAPs) and Shape-Memory Alloys (SMAs) to replicate the way that muscle fibres contract, but these technologies are still mostly in the experimental phase of implementation. But if you want to write a story about androids with human levels of efficiency, saying that they use EAP or SMA based synthetic muscles, is going to be very plausible.
Next, you need to address the efficiency of computers. If you want to really make your robot approach human energy efficiency, part of your design will need to be a whole new kind of computer processor designed specifically for running AIs on much less power than modern computers use. Such proccessessors are already being designed, but as of yet, they are still theoretical. The idea behind them is that modern computers are all designed for linear mathematics which makes them really inefficient at hosting the neural network programs used by AIs; so, if you were to design a whole new kind of computer chip designed for massively parallel processing more like a human brain, you could get your idle efficiency as good, if not better than a human.
The last efficiency issue is the battery, but this is more of a non-issue than most people give it credit for. People like to talk about how much more efficiently bio-chemical energy can be stored than using batteries, but batteries come with some meaningful advantages to. Most notably: they don't need a digestive system to support it. So, going back to our 190lb human, you are looking at about 25-35lb (11-16kg) of digestive organs and feces before you even get to any actual energy storage whereas all an android needs to recharge is some kind of power socket. Add in about 45lb (20kg) of body fat at an average healthy BMI, and a 190lb human body devotes 70-80lb (31-36kg) to its power system.
Once you have a good synthetic muscle to make your android out of, you can eliminate a lot of the wasted weight caused by our current clunky mechanical systems I previously brought up. This means you can focus more internal space/weight on power storage making a 70-80lb (31-36kg) lithium ion battery far more reasonable than in most android designs to date. This gives you about 22-25 kW/hr of power storage meaning that the android could actually perform about 2 days of continuous activity, 10 days of idle activity, or a couple of months fully powered down but with the battery in.