Solution or Workaround
In a very tight zoom-in, it seems as if the cusor has a minimum resolution, not allowing itself to be positioned in a smaller increment.
This is data specific, in particular to the mapextent. PC ARC/INFO represents all of its coordinates in single precision, in accordance to IEEE standard of a 4 byte real number. The value can range from 1.1 E-38 to 3.4 E38. Of the 32 bits allocated for storing the number, 1 bit represents sign, 23 bits represent the mantissa, and 8 represent the exponent. That means that the number of significant digits, regardless of the exponent, is limited to 2 to the 23rd power, or 8388608. This is a little less than seven digits of accuracy for a floating point number. An integer in single precision is about ten digits of accuracy.
In other words, if the maximum x-coordinate is represented by 5 digits, then you would not expect to be able to resolve a distance of 0.05 or less since the five constitutes the seventh digit. This is true regardless of the units. Since the coverage cannot resolve this distance, neither can the cursor in
Perform an x,y shift and thus remove the redundant digits at the beginning of the coordinate value. For instance, if x varies from 10250 to 10500, then 10000 can be subtracted from all the x-coordinates and their position relative to each other has not changed. This is best performed by using the TRANSFORM command. For example, at the Arc prompt:
CREATE new old
CALC XTIC = XTIC - 10000
TRANSFORM old new
Arcview and Workstation ArcInfo support double precision coordinates.