The problem is in the first gain stage and is due to slew limiting in the feedback network. I resolved the problem by replacing 2 x 47pf capacitors (1 per channel) with 22pf. This reduces the amount of local shunt feedback and removes the slewing problems. The 47pf caps are ceramics and (from memory) they are the only 47pf caps on the board so they will be easy to find. I also replaced them with polystyrenes for better performance. After the mods I could hear no ill effects and in fact I felt that if anything there was an overall improvement as well as better transient performance. I noticed the problem more on SACD, where the extra bandwidth was further compromised by the slew limiting. BTW, I was clearly able to see the problem on a CRO with a 10KHz square wave applied. It was evidenced by a massive overshoot on the leading edge. After the mod, the 10KHz waveform was pure. I also submitted my findings and corrections to MF. Unfortunately, they never contacted me or thanked me.
I hope this helps.