CPU is a precious resource that must be shared between the processes concurrently running on a machine at a given time. Some operating systems of the past implemented this sharing using a technique called collaborative multitasking, which is based on the processes' explicitly relinquishing CPU control to the OS so that this control be passed on to other tasks. This can be very inefficient, as sharing cannot be controlled by the operating system, but depends on the good judgement of third party application programmers; furthermore, a faulty process that fails to yield CPU or hangs will collapse the whole system. So, modern OSs implement preemptive multitasking, in which the system scheduler can give or take CPU control without any special help from the processes. Usual implementations of preemptive multitasking group processes into two categories: IO-bound processes, which consume most of the time waiting for user input or dealing with secondary storage, and CPU-bound processes, which take up as much computing power as they are given. Some scheduling algorithms reward the least CPU-hungry processes by giving them higher priority to regain CPU.
In a desktop machine, user attention is as much of a precious shared resource as CPU is: concurrently living applications try to attract the user focus by displaying some visual effects or directly putting themselves on the forefront. In the early, more naïve days, applications could freely jump into the foreground of their own accord, which could quickly degrade into an all-out fight (specially if malware was involved) for the user's attention, or "attention thrashing", and render the computer unusable. This situation is reminiscent of the first candid collaborative multitasking systems; incidentally, attention thrashing led some OS vendors to change the behavior of some of their APIs to try to restrict the applications' freedom to steal focus from each other, though workarounds abound. In the web browser arena, attention thrashing can be even more daunting, and most users end up entirely blocking pop-ups and other forms of focus stealing.
As I see it, user attention should then be explicitly taken care of by the operating system in the same way as CPU control is; in fact, we can learn from the history of multitasking and move from the current "collaborative" situation to system-controlled scenarios where applications request attention and are given it based on attention availability and some measure of application fairness. So, applications that request attention very rarely and do so for justified motives (where this last attribute can be estimated indirectly by, say, recording the amount of time spent by the user on the application when this pops up) are ranked higher and can be given more leeway to call on the user. On the other hand, noisy applications are rapidly ranked low and will not be able to steal focus unless the user explicitly clicks on them.