Steve & Andy have given the key to the solution. It was said the board has a RTC but the oscillator takes 3 sec to stabilise. I doubt if the error during the 3 seconds would stop you from counting how many loops the cpu does in a given timeframe - even if the absolute accuracy of that timeframe may be a little shaky - you're trying to detect a 2:1 difference in frequency. If you have external eeprom (i2c etc) on the board you can store a value to determine the oscillator frequency - you could even have a special first start detection of the frequency when the eeprom has no information and gets it's first time initialisation. Another technique is to measure the write time of the eeprom - it should be within a small tolerance given a particular part/manufacturer - extreme perhaps but it may be a solution.

Also one of the respondents suggested two software versions - that can create untold havoc - say a customer on the other side of the world needs a spare part - which one do you send? Of course everything you ship has a serial number - of course it does and you've logged all this at manufacture or maybe not! Therefore it's a guess as to which software version to send - the guess is normally wrong and the customer is really annoyed and you look like an idiot. Also it means you have two software versions to maintain and test. Best to keep with the one software version where possible.