Unfortunately machine check numbers are highly system specific. I
can hazzard a guess, based on MCHK_K_OS_BUGCHECK as it appears in
the eb164 PALcode: the kernel issued a callsys instruction.
The only way I can think of for this to happen is for the kernel to
somehow get back to user code without leaving kernel mode. This can
happen with stack corruption.
Some versions of the em86 patches cause the unaligned trap handler to do
just this. I suggest you enable DEBUG_MCHECK in arch/alpha/kernel/t2.c,
and see where the machine checks are coming from.
r~