This is given in SQL Server 2000 BOL.
When an instance of SQL Server starts, it typically acquires 8 to 12 MB of memory to complete the initialization process. After the instance has finished initializing, it acquires no more memory until users connect to it and start generating a workload. The instance then keeps acquiring memory as required to support the workload. As more users connect and run queries, SQL Server acquires the additional memory required to support the demand. The instance will keep acquiring memory until it reaches its memory allocation target, it will not free any memory until it reaches the lower limit of the target.
To acquire as much memory as possible without generating excess paging I/O, each instance of SQL Server sets a target of acquiring memory until free physical memory on the computer is in the range of 4 MB to 10 MB. This range was chosen because testing has shown that Windows NT and Windows 2000 have minimal memory swapping until the memory allocations equal the available physical memory minus 4 MB. An instance of SQL Server that is processing a heavy workload keeps the free physical memory at the lower end (4 MB) of the range; an instance that is processing a light workload keeps the free memory at the higher end of the range (10 MB). An instance of SQL Server will vary its target as the workload changes. As more users connect and generate more work, the instance will tend to acquire more memory to keep the available free memory down at the 4 MB limit.
As the workload lightens, the instance will adjust its target towards 10 MB of free space, and will free memory to the operating system. Keeping the amount of free space between 10 MB and 4 MB keeps Windows NT or Windows 2000 from paging excessively, while at the same time allowing SQL Server to have the largest buffer cache possible that will not cause extra swapping."