Applications use memory most efficiently when they reference data in the same sequence or a sequence similar to the order in which the data is stored on disk. This is called locality of reference. When an application needs data, the data page or file is mapped into the cache. When an application's references to the same data, data on same page, or in the same the file, are localized, the data they seek is more likely to be found in the cache.
The nature of the application often dictates the sequence of data references. At other times, factors such as usability become more important in determining sequence. But by localizing references whenever possible, you can improve cache efficiency, minimize the size of process's working set, and improve application performance.
In general, sequential reads, which allow the Cache Manager to predict the application's data needs and to read larger blocks of data into the cache, are most efficient. Reads from the same page or file are almost as efficient. Reads of files dispersed throughout the disk are less efficient and random reads are least efficient.
You can monitor the efficiency of your application's use of the cache by watching the cache counters for copy reads, read aheads, data flushes and lazy writes. Read Aheads usually indicate that an application is reading sequentially, although some application reading patterns may fool the system's prediction logic. When data references are localized, a smaller number of pages are changed, so the lazy writes and data flushes decrease.
Copy read hits (when data sought in the cache is found there) in the 80-90% range are excellent. In general, Data flushes/sec are best kept below 20 per second, but this varies widely with the workload.