How can 8 bytes possibly be enough to hold a task's context in a preemptive environment?

1 PSW
1 ACC
1 B
2 DPTR
2 PC
1 SP

That's 8 bytes right there! You've said you'll use Keil Cx51. The compiler utilizes registers R0-R7 too, you know. I suppose you'd be OK if you only have 4 tasks with 1 register bank dedicated to each task. You would have to have really simple tasks that called no other functions and did not invoke any of the compiler's runtime library functions to avoid any stack usage.

So you've changed from a round-robin scheduler to a prioritized scheduler? There is a wealth of information out there on all types of scheduling algorithms. Need I suggest "Google"?