If rows that are next to each other in an index also sit near each other in the table, then following the index reads a handful of heap pages instead of one page per row, and the engine tracks this relationship as a statistic called correlation. Tables that grow by appending are naturally well correlated on their creation date, and badly correlated on almost everything else. The CLUSTER command physically rewrites a table in one index's order to restore that alignment, which helps range queries on that column a lot, helps nothing else, and drifts back out of order as the table is updated.
CLUSTER rewrites a table so its rows sit on disk in the order of one chosen index. That matters because when matching rows are already next to each other, reading a range through the index touches a …