It depends on the design and the goal, offset is the first problem, but depending on the application it's not a problem, if everything is AC coupled the offset won't upset anyone. If you look Neve DOA with a single input transistor, the NI input is the base, the II is the emitter, so from that you already know about 0.65V DC offset, and the gain at each end is way off, so not a big problem. I'm building a mixer with DOA and I don't want electrolytic caps in my direct signal path, because I only can get the crappy ones, so I put some effort to match transistors, not too much though, just the DMM in hFE and diode for the VBE and that's it, it's not the correct way to do it but my offsets are down 1mV for most of them, which is good enough for the DC servo to be working in a comfortable way. I bought 400 BJT and measure all of them, and make a xls and pick the closer ones, it took me a couple of hours to do all that, controlled room temperature (not controlled but measured within 1ºC in the time I did that) and everything comes out fine. I've already build a small bunch of them with no problems. If you like simulations you could run a couple with perfect match and big mismatch and see the differences, of course the offset would be one but I don't know how bad the distortion would be affected. Another case to have in mind is the 990, which was designed around a perfect match pair, LM394, <50µV IIRC in an IC package which maintain temperature match for them, and is a really nice DOA out there. The 2520 for example, has two LTP at the input and it keeps being differential in some way up to the last stage, but I never really put myself to analyze that circuit, which is not as simple as it looks, it have a couple of tricks, don't know the why of each. If you look at the 990 is much strait forward, LTP at the input, a current buffer, then voltage gain, (some sort of darlington like configuration for this) and then the output stage, the rest is current sources and mirrors.
As I said before, depending on what you are looking you may get away without matching, and really good matching is hard to do and need a lot of transistors to start with, in my bunch of 300 I didn't get a lot of perfect matches, and if you measure correctly to look for the difference between VBE much less than 1mV would be hard to find.
JS
