I don't think there are solutions different than "keep the latest N lines while reading forward the data" or "start from the end and go backwards until you read the Nth line".
The point is that you'd use one or the another based on the context.
The "go to the end and go backwards" is better when tail accesses a random access file, or when the data is small enough to be put on memory.
In this case the runtime is minimized, since you scan the data that has to be outputted (so, it's "optimal")
Your solution (keep the N latest lines) is better when tail is fed with a pipeline or when the data is huge.
In this case, the other solution wastes too much memory, so it is not practical and, in the case the source is slower than tail (which is probable) scanning all the file doesn't matter that much.
与恶龙缠斗过久,自身亦成为恶龙;凝视深渊过久,深渊将回以凝视…