Assuming that we change the L1 allocation order as suggested, another 
trick that would enable us to reduce the number of static L1S needed would 
be to migrate any process using a static L1 to a dynamic L1 as such a 
dynamic process terminated.  Then the static L1s could be used purely for 
handling transient overload and we wouldn't get into the situation where 
over time they all became used by long-term processes.

Just musing and putting this down so that it won't be forgotten.

R.