The biggest reason I can think of of is because you have a system that doesn't tolerate leap seconds, and you've decided to just omit them entirely for safety.
(Of course, time smearing is usually a better option in those cases since it eventually converges with UTC after a leap second occurs.)
Ah, true. I can think of microcontrollers without RTCs. But I usually handle that by ignoring time completely until acquiring it by the data logging device (usually a computer with appropriate time sense).
(Of course, time smearing is usually a better option in those cases since it eventually converges with UTC after a leap second occurs.)