There is one important assumption: Individual rows are written to disk.

Only then Row- vs. Column-orientation makes a difference. (Or more precise block devices which include SDDs as well.)

If no block device is involved, then the argumentation is nil.

Just take an Excel sheet as example: Which way can you write a tupel of data? Starting at cell A1 and then continue with B1, C1,.. or A2, A3,...? It does not make a difference.

This is why in-memory databases are used for cases where OLTP and OLAP is needed and that is the norm. They periodically dump the current memory state to disk to have a persistent savepoint and all follow up transactions are available in the transaction log files, just like in any other database as well.

Savepoint + transaction log lets reconstruct the current state.