Experimental: particles sent to another rank or removed as lost are now dropped from the swarm in one pass instead of one at a time, and each rank's spare swarm capacity is 10% of its own particle count instead of the global count. Previously every removed particle re-zeroed a buffer the size of the whole run's particle count, which made migration dominate the step at large particle counts and gave each rank memory of order the global count. Results are bit-identical, including Brownian runs, because the survivors keep the same order. Cluster scaling evidence is pending.