A new Delta Lake benchmark found that no physical data layout wins every analytics workload. Liquid clustering delivered the lowest latency for point lookups, time-range aggregations and large aggregations, while ZOrder performed best when queries repeatedly used the same selective columns.
The test used 50 million synthetic retail records and compared date partitioning, ZOrder and liquid clustering across four query patterns. In one surprising result, liquid clustering completed point lookups faster despite scanning more bytes. The author attributes that to better locality inside files, which reduced storage seeks.
ZOrder showed the opposite trade-off. It pruned files effectively for stable multi-column filters, but a date placed second or third in its key dispersed sequential values. Time-range queries were 78 percent slower than on a simple date-partitioned table in this benchmark. For an exact matching filter, however, liquid clustering was nearly five times slower and read about ten times more data.
The results come from one controlled synthetic dataset, not a universal guarantee. The practical recommendation is to test the three most important production query patterns on representative data. Stable, narrow predicates can justify ZOrder; diverse or changing access patterns may favor liquid clustering and its lower operational burden.