Numpy log slow for numbers close to 1?
At the time of writing, most chipsets evaluate log using a Taylor series coupled with a table of specific pre-computed values.
With the Taylor series, a number closer to 1 is slower to converge than a number further away from 1. That could go some way to explaining the difference in execution time observed here.
0.99 may also be closer to one of the tabulated values, which would also help.
Or your observations may not even be statistically significant.