The short answer is that you need better control (feedback) to do it. Practically, you will never be able to calibrate the system accurately enough to go straight for more than a few tens of robot-body lengths. Once you dial it perfectly for one set of conditions, the environment or wear conditions will change and you'll have to tune it up again.
Surface conditions, traction, attitude, motor-isolation (the distribution of electrical power to each motor from a common power source), and many other real-time operational factors effect forward velocity for each side of the 'bot.
Depending on your precision requirements, something as simple as a magnetic compass (position as far forward on the robot as possible to maximize its responsiveness) could help you maintain a heading during forward motion.
Often it isn't critically important precisely which direction your 'bot is moving; rather, it simply needs to be making progress on some task (follow the leader, search for a target, etc).
If you post greater detail on your robot and its design objectives, I could assist you further.
A note about magnetic sensor placement
But, why should I "position [the magnetic transducer] as far forward as possble"? Isn't it true that the angle is the same? Yup. That's true, but the magnitude of the Earth's magnetic field is not. You are standing in a different spot on Earth.
Imagine your robot is as big as a car. If you sit in the geometric center of the car and the car pivots about you, your coordinates on the Earth have not changed; only your attitude has. Now if you are sitting on the hood of the car and the car repeats it's previous motion, both your attitude and your coordinates have changed. Changing coordinates produces a bigger difference in magnitude of the Earth's field than rotation alone.
Over the last few years I worked on a team with Dr. Dwight Veihland of Virginia Tech, arguably the world's leading expert on super-high sensitivity magnetic sensors. If I were to crystallize the body of his work (like in this example), I would say that he is always in pursuit of greater signal-to-noise ratios in the detection of ever tinier magnitudes.
Any increase in the magnitude difference you can generate makes life easier for your sensor... and in this case, you get it for free. A number of DARPA grand challenge robots placed the GPS sensor forward for this same reason.