I too have experienced the throttling at some SpC's. I don't yet have enough information to say for sure what the cause is, but I can tell you for sure what it isn't:
1. Car overheating. I can verify, at least in my experience, you can't get the car (battery) hot enough to cause the throttling I've experienced by highway driving regardless of weather. As others have also pointed out, sometimes just a stall change rectifies the issue, proving it's on the SpC side, not the car side.
2. Utility Causes. The limit of the utility interface is the transformer. If the SpC load was to exceed the rating of the transformer for too long, it would blow it's primary fuse and the whole site would go dark. Presently, There is no mechanism for the utility to tell the SpC controllers to throttle. Now this wouldn't stop Tesla from throttling due to demand fees, but AFAIK the demand fees, at least here on PG&E, are going to get triggered the first time someone SpC's one single car. So they likely just have to eat them. This may very from utility to utility, but I can't see them trying to skirt it be throttling.
Theories:
1. Thermal concerns in the cable/connector. There is thermal monitoring, and I suspect they also watch total voltage drop from cabinet to pack. If this is exceeded due to a worn cable or connector, this may cause a given stall to limit it's current. I've definitely found stalls with bad connectors. I try to carefully inspect the connector before I connect it to my car. I use a bright flashlight and look for excessive wear or signs of prior overheating and if I find that, I switch stalls.
2. Defective modules. As others have mentioned, the SpC cabinets are made up of 12 chargers from a pre-refresh Model S. Each one of these is capable of about 12kW by itself, but they are arranged in 4 groups of 3 each for a total of up to 144kW (depending on generation and assuming everything is perfect). The 4 groups can be switched independently from one paired stall to another (A/B). A failure could result in one group being offline which could kill 25% of the cabinets capacity. This means that each cabinet can partition the stalls in the following modes:
A=100% B=0%
A=75% B=25%
A=50% B=50%
A=25% B=75%
A=0% B=100%
Failures of one or more modules or one or more groups could easily limit that cabinets power output.