Provided that "restart" means jump to 0 and then set all registers which can be set explicitly from software, if an ISR has exit through ret rather than reti, the symptomps would be the same - interrupts of the same priority are prevented from being executed as long as *internal* registers (not readable nor writable explicitly by program) indicate that an ISR of the same or higher priority is still being executed.

You might consider posting the extint 1 ISR here for public scrutiny, if it's not too long.

There might be no visible problems there, though; it might be for example also a stack overflow problem (especially if all 4 ISRs might nest in each other, given they are all of different priority).

JW