The high cost of random access to DRAM comes from jumping between rows (blocks of consecutive few -to- few tens of kB). If you read sequentially with gaps, you

The high cost of random access to DRAM comes from jumping between rows (blocks of consecutive few -to- few tens of kB). If you read sequentially with gaps, you only have to open each row at most once. So it's like normal sequential reading, except that you may end up skipping few rows here and there and you may end up not transferring some parts of the rows you opened. So the time to completion can only be shorter, even if average throughput is lower.