There was an error in the bit arithmatic for the can id.
Basically, if the 15th bit of the id was set, then the upper 16bits were 1 less than they should be.
e.g. 0x0001(8-F)XXX -> 0x0000(8-F)XXX
e.g. 0x0000(8-F)XXX -> 0xFFFF(8-F)XXX
The offending operation is (tbufdata[MCP_EID8]<<8). This promotes an unsigned char to a signed int, which results in the corrupted ids. The fix is to cast it to an unsigned int. Note that int's are 16 bit on avr.
woodward1
reeedstudio