Think of it like having two small filing cabinets vs one large cabinet. Even with both perfectly organized, the smaller ones are faster because:

• Each table's data blocks are more compact (fewer columns = smaller blocks)
• Less I/O overhead since you're only loading relevant columns
• Better cache utilization as more rows fit in memory

The tradeoff is maintaining multiple tables and the union view. But when you heavily filter on the sort key, the I/O and memory efficiency gains usually outweigh that overhead.

I've seen 2x performance improvements using this approach, especially with wide tables where queries only needed a subset of columns.