If properly cared for, I would imagine the USB connector would wear out first… if you prototyped with it continuously. If you built a project and committed a Spark Core to it… the hardware would last “forever”. There are no electrolytic caps on the board to dry out. Even if they did have caps, I have 35 - 45 year old electronics full of electrolytic caps that still power up today and run just like they did back in the day. YMMV with electrolytic based on the quality you start with, but here we’re not talking about them anyway 
I realize this is not all very scientific, but an actual MTBF number would be best found with some accelerated life testing and larger sample sets. Also, what are we considering a “failure” in the Spark Core exactly? Data glitch, some API not working, total system failure (i.e. won’t boot)?
Some of the MTBF data can be estimated based on work that the MFG’s of the components of the Spark Core have already done… for instance the CC3000 Wi-Fi module, is something like 4.67*10^8 hours (i.e., “forever”)
http://focus.ti.com/quality/docs/singlesearchresults.tsp?&templateId=5909&navigationId=11213&appType=folders&searchType=orderableOption&partialSearch=false&mtbfType=true&orderablePartNumber=CC3000MOD
The MTBF for a 10V tantalum capacitor on a 5V bus is something well over 1000 years.