@David: it does not work. It makes sense as well. From Nick Gammon's site (Gammon Forum : Electronics : Microprocessors : Interrupts):
However the external interrupts (where you use attachInterrupt) do a bit more, as follows:
...I count 82 cycles there (5.125 µS in total at 16 MHz) as overhead plus whatever is actually done in the supplied interrupt routine. That is, 2.9375 µS before entering your interrupt handler, and another 2.1875 µS after it returns.
Even with clock divided by 128 the bits stream in to quickly. Too bad, because it would have made an elegant solution. I think that if you would use a very slow software SPI on the sending side this might work, but I'm looking for a bit of speed here.
I'm still wondering whether I can optimize the code a bit.
Cheers,
Jack