To expand a little on Peter's post, there is a point in line count where it becomes physically impractical to make the encoder disk grating any finer. I think BEI, for example, stops at about 5000 lines in their size 35 encoders. Anything above that is interpolated from the sine and cosine grating interference signals anyway and the square wave pulses are generated by an ASIC from that interpolation. For high resolution applications why not move the interpolation to the receiver? Both EnDat and Hyperface can be interpolated to well in excess of 1,000,000 counts per revolution.
Also, as the PPR of the encoder increases the pulse frequency at a given speed also increases. Unless you are extremely careful there is a point where the pulse edges start to degrade to the point that they are unreadable, due largely to cable capacitance. The SIN/COS interface is somewhat immune to this. You have the possibility of phase shift affecting the absolute position by some amount due to the same physical effect. But ultimately no position information will be lost.
Keith