NRF24 weirdness - doesn't seem to work reliably with [0,0,0,0,1] and [0,0,0,0,2].

Very simple setup - Two NRF's talking to eachother, connected to a single Espruino. I figure this is a best case scenario, right?

SPI3.setup({sck:B3, miso:B4, mosi:B5}); // B6, 7 used for other pins. 
SPI1.setup({sck:A5, miso:A6, mosi:A7}); // B0, 1 used for other pins.

var nrf1 = require("NRF24L01P").connect( SPI1, B0, B1, 16);
nrf1.init([0,0,0,0,1], [0,0,0,0,2]);

var nrf2 = require("NRF24L01P").connect( SPI3, B6, B7, 16);

nrf2.init([0,0,0,0,2], [0,0,0,0,1]);


var test=[222,173,190,239,222,173,190,239,222,173,190,239,222,173,190,239];
var test2=[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16];

function test1() {
  console.log(nrf1.send(test2));
  while (nrf2.getDataPipe()) {
    console.log(nrf2.getData());
  }
}

function clear() {
  while (nrf1.getDataPipe()) {
    console.log(nrf1.getData());
  }
  while (nrf2.getDataPipe()) {
    console.log(nrf2.getData());
  }
}

Using the addresses above, I get stuff like:

>test1();
true
[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16]
=undefined
>test1();
true
[0,129,1,130,2,131,3,132,4,133,5,134,6,135,7,136]
=undefined
>test1();
true
[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16]
=undefined
>test1();
true
[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16]
=undefined
>test1();
TX not received 82
false
[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16]
[0,129,1,130,2,131,3,132,4,133,5,134,6,135,7,136]
[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16]
=undefined
>test1();
TX not received 82
false
[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16]
[0,129,1,130,2,131,3,132,4,133,5,134,6,135,7,136]
[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16]
=undefined
>test1();
TX not received 82
false
[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16]
[0,129,1,130,2,131,3,132,4,133,5,134,6,135,7,136]
[1,2,3,4,5,6,7,8,9,10,11,12,13,14,15,16]
=undefined
> 

Note here that it worked well for a while, sometimes quite a while at a time without going into fail-every-time mode, but throughout that time, it's occasionally receiving bogus data (!) shifted one bit to the right - which tells you something about the value of the NRF's CRC check...

Well, that's discouraging isn't it?

Except... This works:

nrf1.init([0,0,127,0,1], [0,0,127,0,2]);
nrf2.init([0,0,127,0,2], [0,0,127,0,1]);

In fact, every combination of addresses I've tried so far - except [0,0,0,0,1] and [0,0,0,0,2] - seem to work just fine.

Am I the only one encountering this behavior?