I think Erik's concept of 'real-time' is flawed. There's no doubt that using a pre-emptive RTOS may affect interrupt response time, but by the same token, using a co-operative system (or no 'system' at all) there are times where interrupts are disabled so that certain operations can be atomic. To do anything useful, interrupt code must share variables with non-interrupt code so therefore we need atomic operations which imply we need to disable interrupts (especially on a '51).

Ultimately, what cpu you use, at what clock rate, whether you use C or assembler and whether you use spaghetti code, a RTOS or some form of co-operative tasking all depends on the person that has to make it all work. If that person understands the strengths/weeknesses of the various options, he/she should be able to make a good judgement and have a successful project. This is engineering.