Are you saying there is cache thrashing because callers often sercice rheid own requests themselves? If you don't want to service requests in the same thread, doesnt it mean you have to spend entire core(s) for running the kernel?