The "bayesian regret" is almost as good as RCV, but it's much simpler and doesn't suffer from the misperception you highlighted.
Approval Voting is not problematic for situations where you don't need ballot secrecy and approving is, e.g., a binding commitment to participate if the approved option is selected (or where non-approval is a binding opt-out), or where some other consistent concrete consequence of approval exists.
A similar problem exists with most numerical rating (as opposed to ordinal ranking) methods, and to other ranking methods with smiller number of rankings than alternatives, and, perhaps, to a much lesser extent, to ranking systems that allow (but don't require) equal rankings.
Because most analyses silently assume a consistent the mapping from an internal utility function to ballot marking, this is frequently overlooked.