OK, since no-one knows anything, I've done some playing.
Firstly, yes it works. At least, the ethernet works, I haven't tried the card slot. I've run both the example web server and ntp client and all seems fine. LEDs on the board light up, and so do the LEDs on the jack face - there's a green one to the left that appears to be traffic, and an orange to the right (towards the row of six holes) that appears to be 100M (but I don't have anything other than 100M lying around to test it with). There are small SMD LEDs immediately adjacent on the board that do the same thing. The ones on the board are opposite colours (traffic is orange, 100M is green).
The column of four LEDs by the crystal are link, RX, TX and ON. Green, orange, orange, green.
Reset button works and resets the arduino.
Now for the interesting bit. Numbering the row of six holes as 1-6 from the one nearest the board edge as 1 to the one furthest from the edge as 6 and adopting normal RJ45 colour codes (ie as T568B at ANSI/TIA-568 - Wikipedia), the holes are:
1: blue pair (ie unused for data)
2: brown pair (unused for data)
3: orange pair (a data pair)
4: green pair (a data pair)
There doesn't appear to be any connections to 5 and 6 (you can see that from the board photos - very short traces to two points just under 2mm spacing).
There are no diodes (no rectifier bridges) in the jack, so far as I can make out. I detect no diode drops when I inject on pins 4, 5, 7 or 8 of the jack, and whatever polarity I inject comes out on the pads as above. I think the pads are simply connected to the pins. Presumably on pins 1, 2, 3 & 6 of the socket it's actually to the centre tap of a coil in the jack (since otherwise the ethernet wouldn't work), but I can't detect that. I can't tell whether (for example) 4 & 5 are shorted more directly than 1 &2 (cheap multimeter), so don't know if the non-signal pins have coils across them. Probably not, but that's a guess.
I've had 15V at about 100mA and 3V at about 300mA of power through the pins 4, 5, 7, 8 while traffic flows on the ethernet and it's fine with that. I decided not to actually try and fry it, so stopped at those levels. I've had 50V on the pins while traffic flows (but no current draw) and it was likewise happy. I don't have an injector that will put power on the data pairs (though that's obviously allowed by the standard) and I'm not going to build one just to see if it works - the power injector modules I have just use the non-data pairs anyway.