The PLC manufacturer should have hardware specifications on the accuracy of the clock and, most likely certification of clock accuracy to a certain degree. This is your base accuracy. You then need to evaluate how the timer function operates and how frequently and with what variation the timer function is executed.
For example, a certain model of controllogix may have a timer chip certified by Rockwell or even a 3rd party like TUV to be accurate to plus or minus 10 microseconds. The Logix5000 implementation of the TON timer introduces a rounding error of up to 500 microseconds every time it is called because it is accumulation based instead of start time based. Your cycle is set to cyclic instead of continuous at 5 millisecond cycle time and VxWorks (the runtime or OS that controllogix uses) certifies that the jitter/variation on cycle start time is 5 microseconds or less. You are timing for 300 seconds.
Every cycle then introduces a possible 515 microseconds of accumulated error (154.5ms over 300 seconds), plus you have the 5ms possible worst case error from the cycle time itself. You're looking at around 160ms maximum error for the above example.
Contrast that with all the same numbers but not using an accumulation based TON, instead comparing a start time to the current time to cut out the accumulation error, and you are looking at just 15 microseconds per cycle instead of 515. That gives you a maximum error of 9.5ms for a 300 second timer on a 5ms. Actually, I think I'm accounting for clock accuracy and cycle jitter incorrectly and it would be more like 5.015ms.
You cannot use the continuous cycle if you need to validate timer accuracy.