Training wheels are the longest way around teaching a brain how to control a body that can eventually ride a bike. (There again is that assumed speed metric, faster is better.)
Let's make the assumption that the sooner a person learns how to ride a bike without training wheels the better. Let's also say up front that balance bikes are the best way to do this, and then talk about why.
The brain is a neural network that makes sense of the world by relating new experiences to patterns with which it's already familiar. Psychologist Donald Hebb proposed this conceptual mechanism for how we learn shortly after World War II. As large-scale computer models were first being imagined the brain was the ultimate muse. His Hebbian learning model still stands up today in one form or another.
Without going deep into the mathematical proof--because I can't--the Hebbian model asserts that neurons that fire together are more likely to fire together again. Furthermore, those neurons firing together form cliques that suppress the neurons around them that are not firing.
Like any great fractal, patterns of our socio-emotional existence, or patterns of our immune system, or patterns of our political systems over the ages all perform in similar ways. It can be helpful to think of patterns of learning in this way.
Rising up is, in part, is often relative to the suppression of others.
Back to the bike.
Balance is something that humans learn deeply at a time when the brain is first developing. Therefore a large percentage of our neurons continue to fire together with the very popular clique of balance as we mature. Neurons that fire together are more likely to fire again, and those that didn't fire together get suppressed. Balance-related neural patterns have a better chance of being accepted into the clique than non-balance related patterns.
Pedaling, if it was ever experienced, was likely long after balance was established. Think tricycle. The tricycle brain and the balance brain never had to to talk, so they created a habit of not talking. Since we are put on a tricycle long after we stopped actively learning to walk we shut off the balance brain and reinforced the joy of moving forward without having to balance. Dopamine kicks in and the "no-need-to-balance-while-swiftly-moving-forward" feeling is reinforced over and over again with each pedal stroke.
You see where this is going.
Pedaling is more likely associated with the Hebbian patterns of not balancing than it is with balancing. Introduce training wheels and do the math.
Balance bikes come in with a stacked deck. They ask the brain to walk in a new way. As mentioned earlier, neural patterns for walking excite a huge percentage of the early learning brain. A balance bike is welcomed into the clique as an exciting variation on walking without any suppression by powerful cliques such as pedaling if the child ever rode a tricycle or similar.
In short order this new way of walking on a balance bike gives way to assisted jumping when a child lifts their feet and flies to the next foot fall. The dopamine hit from this joyful experience invites the brain back for more. Sooner than expected the pattern--reinforced by the gyroscopic effect that keeps a moving bike upright anyway--turns into balanced gliding.
This balanced gliding is further supported by a huge percentage of patterns that were created by early walking and the brain makes this activity 'second nature' without much objection.
With the balance part sorted out and subconsciously unassailable, pedals can be introduced.
Muscle memory of any sort works the same way. It's a layering of the unfamiliar onto a solid base of something familiar. Even if it seems psycho-somatic.
For example, if you tell yourself that making your bed is "exercise", then Hebbian patterns will connect it to the benefits of exercise rather than relegating it to the negative patterns that you might have established around "chores."
Confirmation bias works the exact same way.
Ooops. Was that out loud?