For audio performance testing my personal "model" would be something like a Tektronix AA501, 100K input, full differential with high CMRR, 100kHz bandwidth or greater, capability to analyze signals as high as 30Vrms (The AA501 could deal with 200V!) and down to somewhere around -40dBv. I don't expect to pull that off with an audio interface without external attenuators of course. For output, be able to generate signals up to 30Vrms (I know that can't happen without an external amp), differential or single ended, with variable output that can be calibrated for 0dBFS at minimum of -40dBv, and up to 30Vrms. That would pretty much cover every application, and I know that's not going to happen in one box, but I'm trying to see how close I can come without breaking the bank. I might use the high Z inputs on the M2, but certainly not for everything. I'm not using REW purely as a scope, I have a real scope with 10X probe for that if needed. Its the signal analysis functions that I'd use it for.
One problem I'm seeing with lower cost interfaces is that calibrating input and output so that 0dBFS doesn't clip is not always possible. I have one low-cost interface that has some sort of gain ahead of the ADC that severely limits its maximum input level, and several interfaces that won't actually meet their maximum output spec without clipping. Not that it's a huge problem, but you have to work with your brain fully switched on. It's always surprising to see an interface that states it's "reference level" is +4dBu, but then t's maximum level is +13 or so.