How? All Core-M are made with same production process, have the same processor architecture and the same TDP (up to 6Watt). They only have different clocks speeds and turbo states. Apple confirmed to a german techblog, that their Core-M processors are even limited to a custom TDP of 5 Watt.(
http://www.golem.de/news/macbook-12-teurer-prozessor-kaum-schneller-1504-113366.html).
I think that Speedstepping or switching turbo states is a rather complex task. The higher clocked model is faster for short periods of loads, because it can take advantage of thermal intertia (i hope it's the right term, i am not a physicist nor a native english speaker). When the processor uses its turbo mode, the temperature rises slowly. So it takes a few seconds until it has to reduce its frequency and voltage. In the meantime a task like sharpening a picture is probably done.
If you apply sustained loads there is a downside. The processor will heat up other components like the mainboard, chassis etc. more quickly. This effect influences the processors, so it can't make uses of it's turbo clocks anymore and will reduce its speed. And it probably takes a little bit longer to cool down. The lower clocked model reaches this state a bit later than the higher clocked model, because its turbo is slower. And therefore it can keep up its speed a bit longer and produces the same or slightly better results sometimes. Look at the temperatures and average speed from the french test (started on both machines at the same time during iMovie Export):
Image
I am not saying that Core-M is a crappy processor or underpowered. It's actually very advanced engineering and probably sufficient for most users. But in my opinion old thumb rules (higher clocks = always faster) or geekbench results only tell one side of the story. There is a certain variety in real life tests due to heat sensitivity of the Core M. I couldn't find equivalent in depth english article, where the Core M is tested in the same way. So my posts is based on this article from the german techblog golem:
http://www.golem.de/news/core-m-5y70-im-test-vom-turbo-zur-vollbremsung-1411-110529.html