Had a little time this morning to do some more testing. By biasing both gates slightly off-centre, I was able to reduce/control current in the inverter and remove the source resistors. However, transconductance was too high and clipping was harsh (no subtle compression as the output nears the rails). Pity, as the source resistors definitely contribute towards distortion of the wave form and, after a little testing, I think I know what's going on.
Due to negative feedback, neither the n or p device acts as a follower. Instead, each source resistor forms a divider with the channel of its respective device. The effective resistance of each device's channel changes with the voltage on the gates. Considering (for convenience) the n-channel's source only, the source will drop as the output drops. However, with negative swings on the gates (positive output swings) the n-channel narrows, its effective resistance increases and it pushes the (comparatively small) source resistor down. End result is full-wave rectification, albeit around a shifting DC centre-point. The latter is partly due to inverters making poor amplifiers. The virtual-earth effect isn't perfect so a tiny voltage is seen at each source (it acts as a slight follower). When either the p or n device are pushed into saturation, negative feedback is lost and a slight voltage builds up on the gates, in turn increasing the slight 'follower signal' seen at the source.
Unfortunately, all of the above causes slight distortion to the peaks of the signals. Perhaps this accounts for the slight 'rattle' I mentioned? Bypassing the source resistors cures the problem but creates another: transconductance is too high and clipping is hard (again, there is no soft compression as the output approaches the rails).
Looks like drain resistors might be the way to go (for me)...