Instead of thread-pools for each server/context, use a server-wide top-level threadpool, which is more efficient.