you had to lower the bias on the smtp version.
It's more like the TH version can have much higher bias if needed, esp. with taller heatsinks, while in the SMT version is limited by the heatsinking provided by the PCB itself. Stock TH runs at about 50mA (with 100 ohm resistors), while SMT runs at about 25mA.
In principle, the lower biased output stage would leave Class A sooner with large signal peaks
and a low-impedance (32Ω) load. Class A ends at twice the quiescent current, so for the SMT board with its 25mA, that would be 50mA, or 80mW into 32Ω. With a typical can sensitivity of 100dB/mW and a normal listening level, that would happen from time to time, but not very frequently. However, Omicron stays very, very linear even when that happens (see the bottom of
post 11).
In practice, I find it hard to hear any difference even with my 32Ω Grado GS1000i.
You have to be careful with SMT parts
That's true, and the BOM includes all the right stuff. Only thin film resistors and NP0/C0G (white-blue, not brown) capacitors in the amp proper and MKP caps in the crossfeed circuit. The protection circuit does use brown 1uF X7R caps, but these don't matter. Also, quality, low(ish) ESR electrolytic caps are recommended for both TH and SMT versions - but nothing exotic; the prototype uses Panasonic FC throughout.