Just like a number of (unrelated) algorithms, the sweet spot can be found in the middle between two (or more) optimal solutions.

Precision is sometimes useless due to time and computing resources constraints. Speed can come at the cost of poor results.

I think the answer depends upon the specific use case and environment.