The linear estimation problem constitutes one of the simplest classes of problems, to which the unifying and powerful partitioning approach of Lainiotis applies. In the Lainiotis and Andrisani partitioning algorithm, model partitioning consists simply of decomposing both the initial state and process noise vectors into the sum of two independent random vectors. In this paper, a comparative computational analysis with respect to computer time and storage requirements is made between a new partitioning linear estimation algorithm and the Kalman filter for time varying models. The new algorithm constitutes a per-sample version of the above Lainiotis and Andrisani partitioning algorithm. The practical usefulness of the new algorithm is demonstrated by applying the comparative computational analysis to the important class of multisensor estimation problems.