"it will be inefficient when b is larger in size"... actually the concept itself is inefficient. Any method that you find that has to remove some elements from a large array and then reallocate that array to new contiguous memory is going to be inefficient. It is the changing of arrays that is inefficient, so the solution is don't change arrays (unless you really need to).
It is likely that the fastest and most efficient solution would be to keep the data array unchanged, and simply use logical indexing to refer to the array locations that you need to work with.
