The presumption is that you are talking about electrical load (MW) and not reactive load (KVAr or power factor).
660 MW is a pretty "substantial" machine. Depending on the size of the grid with which it might be connected, it's conceivable that even a +/- 10 MW swing might have an appreciable affect on grid frequency depending on the size of the unit in relation to others on the grid.
One would think that with a relatively stable grid (frequency) and a relatively stable source of steam supply (pressure and flow) and no extraction/admission (we don't know if the unit has any extractions or any admission valves downstream of the main control valve) problems hat it should be possible to have a very stable power output (+/- 1 or 2 MW). We don't even know how many stages the unit has (HP, IP, LP).
It would also depend on how the power measurement is being observed, using an analog meter or a digital meter and with what degree of accuracy.
It would seem you should be able to pinpoint the cause of any instability with some degree of certainty, such as hydraulic supply pressure/flow problems which could be alleviated with hydraulic accumulators. Or leaky hydraulic actuators. Or unstable extraction control valves. Or an unstable grid frequency. Or steam supply issues. Or control valve feedback problems. Or control valve tuning issues (gain, PID loop tuning, etc.).
When did this problem start? How long has it been occurring? Has it been getting progressively worse? What has been done to try to identify the source of the instability?
There are sure to be some flames about this, but, this is just one person's opinion given the information provided. I would say, and this is very subjective, the "allowable range" is anything the unit's owner or the grid regulator will tolerate. Everyone's definition of "stable" or "allowable" is slightly different, if we're splitting hairs. But, this is kind of an odd question given the information provided and all of the possibilities (and only a few of them were mentioned).
I visited a site once where the only change made to the plant was the addition of a digital wattmeter, with two decimal places of accuracy. The Plant Operations Manager was complaining because the digital meter showed a +/- 0.6 MW swing at worst(for a total of 1.2 MW peak-to-peak) on a 40 MW unit, and the swings were typically much less than this even, depending on the load.
The previously existing analog MW meter (with two MW gradations on the scale) was still in use on the panelboard, and the worst "swings" were barely visible on the meter. Everyone at the plant agreed the existing analog meter wasn't moving any more or less than it had always been moving (very slight oscillations that were barely noticeable even at the worst swings indicated on the digital meter), yet the Plant Operations Manager was insisting the new digital meter had introduced some instability into the turbine governor and wanted it tuned out.
The instability was traced to a worn actuator cylinder which was passing a significant amount of hydraulic oil at a particular operating position, where the "swing" was greatest, and had been so for quite some time, and of which the Plant Operations Manager had been long aware. The actuator was not replaced nor refurbished at that time.