If the number of steps "lost" had been varying a little bit, I would have assumed that there had been some slack in the system, so if the motor is moving the in one direction, and you then switch direction, a number of steps will be required to take up the slack before the motor finally starts to drive the mechanics in the different direction.

Another thing to consider - do you turn on/off the power supply to the motor? If you do, then your first steps could potentially be sent out without any power being available to drive the coils. And about 95 steps later, the power supply finally activates the power output after having validated the voltages (let's call it an internal "self-test").