The confusion here is that this is not a pair of transformers, but a single transformer with four windings. This is evident from the physical construction, which has all of the coils formed around a single ferrite core. Two of those windings are connected to form a single center-tapped winding, while the other two are not electrically connected. Thus, the pairing of the coils doesn't actually matter (except for the two that are connected), because they're all part of the same magnetic circuit, and there's a 1:1 winding ratio.
Normally, each Ethernet pair uses a transformer with two center-tapped windings. I'm not sure why they've constructed the transformer this way. If pins 3 & 4 were combined you'd have the traditional pair of center-tapped windings and the footprint would have 180° symmetry, which would be convenient. But I suppose this way gives some flexibility in which way round the isolated pair is combined which allows some flexibility in how the primary and secondary terminals are routed to the PHY and to the outside world. Perhaps there are some PHYs that are designed to drive an independent pair of coils instead of a center tapped coil. As it is, you will need to use two or four of these transformers with pins 3&4 bonded to form the second center-tapped winding on each to create a full set of magnetics for a 100Mb or 1Gb terminal. You may also want to add common mode chokes to each pair for EMI; on pre-packaged LAN transformers and magjacks these are sometimes included. These all-in-one transformers are much more convenient, but of course using discrete components like the SM453229 provides more routing flexibility and may result in a smaller overall device--it's no coincidence that routing flexibility is a selling point of this part!
