After a week and a half of random poking and prodding at different parts of the CMI-01A "Channel Card" emulation for the Fairlight CMI IIx driver in MAME, I finally managed to isolate and fix the sample issue that was affecting ITCH.RS and MILLION.RS.
Without getting too far into the gory details, it turns out that the Motorola 6840 Programmable Timer Module (PTM), one of which exists on each Channel Card and one of which exists on the Master Card to drive primary tempo-keeping, ticks on the falling clock edge when clocked externally, not a rising edge.
For certain particularly long samples, this resulted in the sample either being far too short, or far too long; in this sequence in particular, the "JETTKICK" and "SEXSNAR2" samples were originally being cut off too early. This is now fixed.
27