That's just a buffer to drive the output instead of the chip direct. I'm not a fan of the schematic as drawn and think that the route you've chosen is probably more robust (and offers galvanic isolation, a good thing here). As you say, easy and cheap to play with (even maplin stocks BF245C jfets, its a very useful part)
If you wanted to add a buffer, you could use that circuit ahead of your transformer - but I suspect without a fast scope, willingness to drive yourself mad and knowing what you are looking for, it'll only change things - not necessarily improve at all. For example - since you are buffering a squarewave with a datarate of 2.8Mhz (and given is a square wave you want, that means lots of harmonics out beyond 10x even 20x that) the potential to make a bit of a mess at RF is definitely there. The phono plug itself becomes awkward at this bandwidth- and it is actually part of the spdif standard..!
tl;dr: Easy to get a different perceptual result, but non-trivial as casual DIY to get 'right' I suspect.
ETA:
If you do try the sketch - for heavens sake add a local decoupling cap e.g. physically-small 47-100nF ceramic or film cap from the drain (top terminal) of the jfet to the 0v connection (bottom of the 75ohm reistor, which should be the point yout signal output 0v returns to - and I'dd add a small RF choke on the supply side between the drain and th 9v -15v + ve supply (again, even maplin stocks these, 100uh or so should be fine, value uncritical)