Anonymous Monk is mistaken. This isn't a precision issue. It's banker's rounding.

printf("%.0f\n", 100.5);  # 100
printf("%.0f\n", 101.5);  # 102
printf("%.0f\n", 102.5);  # 102
printf("%.0f\n", 103.5);  # 104
printf("%.0f\n", 104.5);  # 104
printf("%.0f\n", 105.5);  # 106
[download]

Note that all those numbers (but not 100.55) can be represented precisely in floats (0.5 = 2^-1).