When a Spark application is querying data from Azure Cosmos DB, which configuration option has a significant impact on performance due to the physical distribution of data within Cosmos DB?