Recursive Least Squares (RLS) by its structure reestimates coefficients iteratively utilizing one new observation in each iteration. Is it possible to use $n$ new observations in one iteration to reestimate coefficients? Perhaps some similar to RLS method can handle it?
Of course, executing $n$ RLS steps (iterations) sequentially gives the result, but if $n$ is relatively big, utilizing a chunk of observations in one iteration can be more computationally efficient.