Are you splitting the render up across multiple machines?
Try submitting the job with "XSI Batchrender" as 1 subjob, and set the 'Range Execution" mode to 1 partition, so that you get 1 agenda item (1 "frame") that is comprised of the entire frame range, as in "1-45".
That way, 1 machine will pick up the job, and solve the mesh for frames 1-45 in one continuous pass.