Recently, I've read Hennie's Paper. I understood the construction of buffer zones, but why can't it be applied to yield a single-taped Turing machine?
Just because that UTM itself requires $2$ tapes to do its task that efficient. Removing one tape, the magic is gone.
But assuming you have the fantasy power to do efficient $1$-tape simulation, what gains do we have then?
That should be the question rather than your question.