MF10 did the trick. The frequency is too slow to be dual-scoped, hence the weird second image, but hopfully you can see the phase-shift.
Theoretically, gain can be increased by reducing in value the 20Ks to pins 4 and 17 of the MF10: anything down to 10K. Might cause clipping with a 9V supply, though, so I kept things to 2V peak-to-peak.
Data sheet didn't have any info on the input bias current of the MF10's input amp. Might be possible to sidestep the op-amp followers and stick in high value resistors (IM input resistors). Shouldn't cause too much distortion, and would do away with a chip. Also isn't clear whether the 10Ks from 2-4 and 17-19 are necessary. When I could be bothered I'll pull them and see what happens.
Waveform is smooth, although it has a 'thickened'/'serrated' look to it. The data sheet said this would happen. IIR, Penfold uses a small value cap to smooth this out. Probably not necessary (good enough for our purposes).
Clock bleedthrough might be a bit of a problem...a PCB designer's nightmare?
Although a relatively simple circuit the parts count is mainly chips, which is a pain. Phased sines are not something that can be done easily in the analogue domain (it seems). The outputs are amplitude stable, though, and the circuit has reasonably
low current consumption (unfortunately the MF10 can hog around 8mA or so, IIR).
The schematic should be checked against data sheets/CMOS cookbook. Can't guarantee I didn't make a mistake when drawing it.
Well that's my entry. Do I win a cream cake?