We had this issue. When we had this issue, we were running four foreground
method servers with 2 GB each. We then added more physical memory to the
server and allocated 10 GB for each method server and now we are not seeing
this issue, atleast not as often. We are on 64 bit Windows systems, hence
we could allocate this much memory for each method server.