Simple: Not at all. The 68881/2 FPUs don't know quad words (Motorola expression for 64-bit integers). You could convert the arguments into the floating point domain, do an FMUL there and then convert back (with loss of precision, probably).
Don't look that far: Starting with the 68020, the M68k CPUs support a MULS.L operation which does exactly what is asked for: 32x32=64 bits multiplication - With just a slight increase in used cycles over the plain 68000.