This is due to the setup, being specifically designed to do so. Most importantly, he differential settings.
This is one object with two forces applied to it at different points, one greater than the other:
^ <--- force
|
^ |
| |
. <-- center of gravity
What will happen is that the object will move in the direction of the forces, yes, but rotate counter-clockwise at the same time. This is because center of gravity is not on the same axis as either of the forces. Both of them create a torque on the object but the one on the right is greater, rotating the object CCW.
Now to apply this to a car, you have to know how to calculate the torque exerted on an object by a force. This is done by measuring the perpendicular distance between the center of gravity and the axis on which the force is, then multiplying it by the force.
Now imagine your wheels having turned, with the inside wheel at a slightly greater angle (for one there's ackerman steering, for another the contact patch twists more on the inside). Then "draw" infinate lines through both wheels. You will see the inside wheel's perpendicular distance will be much smaller than that of the outside wheel. Combine that with a locked or very stiff differential (high locking factor) and you have more torque being created by the outside wheel than the inside one. Then you say "But the rear of the car will resist this". Yes, true. But this is solved with a rather loose back end, meaning that the car will not resist.
So now you know what to do - unlock/soften the differential (if its a clutch pack, make sure you are editing the power setting, not the coast setting), adjust roll bars (more at the front, less at the rear) and that should fix it. Then, of course you have an understeering beast of a thing to drive that will probably be slower.