When people say "It's a myth that a microwave heats from the inside out", they are trying to correct the impression that heating starts in the very middle of a large item like a roast, and works its way out. For something thin or narrow like a stick of butter or a single serving of pasta, pretty much the whole item consists of "outer layer", so heating is fairly uniform throughout except for variation caused by the microwave variability (which means that it is actually not uniform at all).
The answer by g s explained why the most melting would be between the two ends of the stick, but I have the impression that you are seeing the interior melt before the top surface. What I suspect is happening is something like this:
- Somewhere along the stick is a region of maximum amplitude
- In that region, there may be an amplitude peak inside the stick, but heating may be fairly uniform on that small scale of 2-3 cm.
- If there is an amplitude peak inside the stick, then that will explain your observations. But also I am not sure that g s is right about air cooling being irrelevant. Plus there is radiant cooling. (And evaporative cooling, maybe most important of all, props to user21820.) Heat deposited in the interior of the stick cannot dissipate as quickly as heat deposited on the exterior. Considering how close the stick is to melting already, it would not be surprising for the inside to get slightly hotter than the outside and melt first.
Added later: I did some research. If you google "microwave penetration depth" you will find different sites giving reasonably consistent values. For water at room temperature it's around 1 to 1.5 cm, but it's more for cooked meat.
It's also important to note that there is no sharp cutoff. The penetration depth is defined as the depth at which the power is $1/e$ of the level at the surface, or about 37%. If you go the same distance further in, it will be 37% of 37%, and so on.
Also, the more efficiently it heats, the shallower it penetrates. An analogy might be carrying a plate of hors d'oeuvres into a large crowded party. Water is like a hungry crowd that takes snacks quickly (but still takes fewer if they see the plate is getting empty, so you get that exponential decay). So you don't make it far into the crowd, but they get filled up quicker, just like water heats up quicker. Meat is like people who are too busy talking so not as many people grab the snacks. You make it farther in but the crowd is less satiated. And glass is like a room full of kids who can't even see what's on the tray as you heartlessly walk all the way across the room.
Butter is a mixture of water and fat, so the penetration depth will probably be greater than 1.5cm, but I don't know what to guess.