Classically, disk saturation (thrashing caused by page faults or long disk queues) are the main cause of poor server performance. Even in the most intelligently designed systems, the recommended standard tasks (compact, updall, stats etc) are culprits for high disk usage.
Intrinsic fo the philosophies of running high-intensity tasks (scheduled agents, system processes) that is demonstrated in the documentation, is the basic limitation that some servers/applications are 24-hr in nature.
Even on such systems, compact, updall etc are important, however are LESS important than server responsiveness (eg a global SAAS web application).
Given that disk (and sometimes processor) bottlenecks are the major cause of service degradation, it would be NICE to be able to prioritise tasks. EG, in the SAAS example, nHTTP would be given priority and / or, background "backburner' tasks such as compact should be able to be throttled at certain thresholds (eg less timeslices when certain thresholds are reached by the overall system).