I have just finished two ALEPH3 channel's (had a small problem, see aleph3 problem thread) and found the following:
Meticulous matching pays off. I have my output FETS matched within 0.01 Volt. Try to keep conditions as similar as possible when doing the matching. When the FET is in the testcircuit it tends to heat up and Vgs starts drifting. You can see this if you measure a FET for 5 minutes. Write down the measured Vgs every minute.
If you decide to swap FETS, swap them in pairs. So Q106 and Q107 in one swap or Q108 and Q109 in another.
I have also matched the 5W source resistors using a milliohm meter. I found the spread in these quite large.
The ones I use are UTM 0.47 Ohm +/- 10% tolerance. When I measured these it looked more like +/- 20% tolerance
Voltage across R120..R123 in my first channel is now
0.49, 0.50, 0.49, 0.49 V respectivly. In the second channel voltages are 0.53, 0.53, 0.54, 0.53 V. All measurement taken after an hour playtime for the channel, so speaker and CD connected.
I also checked the temperature for each transistor: 51C (degrees centigrade) where sink temperature was 39C and ambient temperature was 22C.
It goes without saying that proper mounting of the FETS is essential, specially in any class A design with paralled output devices. When there is imperfect mounting differences like you describe could occur. But then this is easily verified with a temperature meter. Fixing it will probably be a little more difficult.
Please do not take your measurements when the ALEPH is cold. You will get funny results.
Hope this helps a little