OK, I did some more research.
Watches/clocks uses something called inhibition compensation which means that at manufacturing the crystal is made to run too fast. Then at the factory the clock is compared to an accurate time source and calibrated in software to be accurate. According to wikipedia a typical quartz clock or wristwatch will gain or lose 15 seconds per month, so +/- 3 minutes per year.
High end quartz watches can be calibrated and certified by COSC and then they have to drift less than +/- 25 seconds per year.
Modern PLCs are microcontrollers and they have real time clock (RTC) components either on the chip or in an additional circuit. According to info from Atmel (one of the large manufacturers) more accurate RTCs have temperature compensation in software. If they are calibrated at room temperature and then temperature compensated in software an accuracy of +/- 10 ppm from 5C/40F to 65C/150F can be expected. +/- 10 ppm is +/- 5 minutes per year.
It looks like most PLC manufacturers don't specify the accuracy of their RTC but some do.
Micrologix 1100
+11 to -30 minutes per year (over entire operational temp)
Schneider Premium PLC
+11 to -53 minutes per year (over entire operational temp)
In general the crystal oscillators are very temperature dependent.
From the data above I can only draw the conclusion that PLCs have less sophisticated RTC circuits and probably no calibration, compared to more advanced circuits that are available and could be used. Even cheap watches have more advanced circuitry /algorithms. To some degree it makes sense because a watch, even a cheap one, is a time keeping device, while a PLC is an automation controller. This is also my personal experience - computers and PLCs are worse than a cheap watch.
Because of the temperature dependency it also makes sense that the drift could vary widely not only from unit to unit but also based on it's environmental conditions. And the same unit could be running to fast at same temperature and then to slow at another. As the crystal ages it's characteristics will change. Of course with some luck and stable temperature some system may not drift as much as others.
Quick rule of thumb is that 1 ppm equals 1/2 minute per year.
Some sources:
http://www.atmel.com/Images/Atmel-42251-RTC-Calibration-and-Compensation_AP-Note_AT03155.pdf
https://en.wikipedia.org/wiki/Quartz_clock#Accuracy
.