Al, you're exactly on the right path regarding the performance of input vs. output transformers. Put another way, output transformers require tighter primary-to-secondary power coupling to maintain low output impedance, and the resulting low distortion and low noise figure. A side-effect of this is the necessity of a larger core, and higher leakage reactances between the primary and secondary. Input transformers on the other hand need to transfer very little power, and so can make effective use of Faraday shields and have lower leakage reactances, at the cost of the requirement of needing carefully controlled secondary impedances for good performance. But in both cases, the distribution of these reactances across the two windings can be controlled in the design of the transformer, and frequently an input transformer will work best with its secondary grounded on one side, or the primary with an output transformer.
A huge complicating factor is the fact that the design and performance of "balanced" inputs varies wildly in high-end audio . . . I would divide them into two "worlds", depending on whether the circuit after the balanced input is balanced differential, or conventional unbalanced. Both have myriad potential design issues.
The main issue with balanced-line-to-differential-circuit input stages is that most of them really offer no common-mode rejection at all, that is, a common-mode voltage on the input translates into a common-mode voltage on the output . . . hopefully (but not always) the common-mode voltage gain is less than the differential mode. The result is that any tiny gain or impedance imbalance within the equipment or cables (and in the following equipment, if it's of similar design) will result in the common-mode (noise) voltage becoming differential-mode (signal) voltage. It's also frequently more suceptible to RF interference than an unbalanced input (there are two input paths), and under no circumstances will the circuit work correctly if fed from an unbalanced source. An input transformer can improve things tremendously on all fronts.
The problem with balanced-line-to-unbalanced-circuit input stages is usually that many of the simpler designs have an impedance balance that's maintained by the open-loop gain of the input circuit, and the critical matching of resistors and circuit trace capacitances . . . and since this is never perfect, the CMRR is poor and usually falls as frequency increases. This can be improved by the buffered "instrumentation opamp" topology, but still all of these approaches almost always result in increased noise over an unbalanced input, as a result of the Johnson noise in the resistors forming the differential subtraction. Here again, a high-quality input transformer almost always performs better, especially because input RFI networks aren't required.
When interfacing with source imbalances or an unbalanced output, CMRR is determined by the ratio of the differential input impedance to the common-mode impedance. In the overwhelming majority of both types of input stages, the common-mode impedance is one-quarter that of the differential-mode impedance, making the impedance balance VERY critical, with very little noise rejection from an unbalanced source. There are two ways of dealing with this . . . raise the common-mode impedance, or lower the differential-mode (signal) impedance. Atmasphere advocates the latter with a 600 ohm terminating resistor . . . the obvious disadvantage is that the overwhelming majority of equipment on the market will perform more poorly into the lower impedance load.
Transformers do the opposite, they raise the common-mode impedance . . . which is why they still work well from an unbalanced source. Input transformers will generally have a higher common-mode impedance than output transformers as a result of the lower leakage reactances mentioned above.
