This is a matter of billing, if Lambda charged per millisecond instead of the current model of 100ms chunks, this wouldn't be an issue.

At scale/high performance there's a lot of waste with the current pricing model.