Working time estimation is known to be helpful for allowing crowd workers to select lucrative microtasks. We previously proposed a machine learning method for estimating the working times of microtasks, but a practical evaluation was not possible because it was unclear what errors would be problematic for workers across diferent scales of microtask working times. In this study, we formulate MicroLapse, a function that expresses a maximal error in working time prediction that workers can accept for a given working time length. We collected 60, 760 survey answers from 660 Amazon Mechanical Turk workers to formulate MicroLapse. Our evaluation of our previous method based on MicroLapse demonstrated that our working time prediction method was fairly successful for shorter microtasks, which could not have been concluded in our previous paper.