Hmmm... Odd-order terms should not increase with a normal differential design, but even-order terms should be cancelled as
@solderdude said. For a typical single-ended design the second harmonic dominates with the third next at a somewhat lower level. Going differential with all else equal (it never is, of course) should cancel the second-order (and all higher even-order) term(s), leaving the third and all the rest of the odd harmonics. In practice there are still some even harmonics left, at a much lower level, and if the bias is different then the odd-order terms may be higher (it is not uncommon to reduce the bias current going differential since the distortion is intrinsically less). Many components use a combination of differential and single-ended circuits inside so you have to know the circuit to predict the performance.
Many low-cost consumer and entry-level pro gear simply uses a resistor on the (-) lead to connect to the shield ("earth ground") so the signal looks differentially balanced. That usually drops the CMRR (common-mode rejection ratio) somewhat because over frequency the driven side has an impedance that is not purely resistive -- it varies with frequency due to feedback and the output impedance of the driver, plus the receiver's input also varies with frequency. In practice, with good design, I have measured 3 to 6 dB lower CMRR than if the connection was fully differential when the driver and receiver were both differential circuits, and maybe 20 dB worse (though still good -- see below) for less robust circuit designs.
Now for the anecdotal stories based on my prior experience. The best differential circuits I have designed and measured (my own or other's), outside of using some exotic tricks to hit 120 dB+, were around 80 dB CMRR for audio products. Going to quasi-differential using active devices usually drops to about 60 dB, not bad, but much lower than fully differential. It depends on how well the (+) and (-) sides are matched, of course. That is still plenty for most applications, and of course much better than a regular single-ended design. And there are active devices (op-amps) that are essentially a pair of opamps with a trimmed resistor network that get back to those 80+ dB numbers. Using resistors to match impedances I have measured about 40 to 60 dB in the past (last time I measured those was several years ago whilst setting up our church's system; a 32-channel Mackie mixer at the back of our sanctuary was pushing 60 dB, and a little Behringer unit we used on stage about 40 dB).
Single-ended inputs can have fairly high CMRR, again depending upon implementation, but in a high-noise environment you have to ensure the input signal (+) and ground are tracking together. Noise coupled into the shield upsets that, and is one reason that adding an extra, isolated shield to a long RCA (single-ended) cable can help. If you have a good ground for the outer shield so it shunts EMI away from the inner shield then the signal itself never "sees" the noise. That is one reason for quad cables -- the extra shield layers allow external noise to be shunted away without noise current circulating through the signal return (signal ground, signal shield) -- it flows through the outer shield layers.
FWIWFM - Don (Now the blizzard warning expired, need to go out and start clearing the 12+ inches of snow blasting around the house and drive so we can get out tomorrow!)