Fried another two programmers :-(

  1. Have you kept the ground wire from the PSU to the H-bridge separate from the ground wire to the ICSP header?

  2. When programming, is there a ground loop? i.e. the programmer is grounded through the PC USB port, and your bench power supply may also have its negative terminal grounded. If so, when the H-bridge switches, there will substantial ground loop currents (especially if you don't have those capacitors I mentioned close to the H-bridge mosfets), and that may be what's killing the programmers. You could try connecting 100 ohm resistors in series with all the ICSP header pins (including ground) and not connecting the +5V ICSP header pin.