By "DC convertor" JC means the rectifier circuit, commonly known as the "envelope detector" or "sidechain". This is the part of the circuit that averages out the audio signal and produces a unipolar (i.e., only extends in one direction from ground or Vref, not above and below ground like an AC audio signal) voltage that essentially represents how loud the signal is at any point in time.
You will see that after the output of the 3080, there is a transistor with 10k resistors on the emitter and collector. This transistor is being used as what is called a "phase-splitter". Two equal-amplitude copies of the signal are available at its emitter and collector. Each "copy" passes through a network involving a cap, a diode, a resistor, and finally a transistor, resulting in an two separate "averages" that get merged when they are combined at the base pin of that C1849 over at the far right.
From an envelope detection perspective, this is a good approach, and thread after thread here will reinforce the notion that "full-wave rectification" (where both positive and negative half-cycles are included in determining the rectified "average") is better in many ways than "half-wave rectification" (where only the positive OR negative half-cycle is used, ignoring the other one).
What JC has noted, however, and I find this quite interesting, is that these two paths are not quite identical. In fact the emitter side of that first post-3080 transistor provides BOTH one of those feeds for the rectifier/envelope-detection AND the audio output too.
Ignore most of the rest of the circuit and imagine, if you will, that this first post-3080 transistor was a single bipolar transistor booster, similar to an LPB-1, and configured as an "emitter follower". Hanging off the output of that transistor is a diode to ground (can you say "clipping circuit"?) and the base-to-emitter path (which is also a kind of diode). It's through a .01uf cap, granted, but it IS a clipping diode pair to ground on the output path. So, what JC suggests for "hi-fi" operation is to simply remove that adjoining cap such that the path from emitter to output pot is isolated from that diode and adjoining network.
Is this optimal? Tough to say. JC knows his stuff, so I trust his judgment on a lot of things (Get his Inside Fender and Marshall Amps book; it's a honey). On the other hand, while taking those diodes out of there may reduce some risk of clipping on peaks, it is well-accepted fact that using half-wave rectification (which is what you'd be left with if the cap was lifted) leads to irritating "envelope ripple", particularly on lower notes and particularly during the decay phase of held notes. If the Dynacomp was set up for a fast recovery (and look around for the "recovery mod" for MXR/Ross compressors), it would be less of an issue since the ripple would be less apparent. A stock Dynacomp, however, is set up for longish sustain, which increases the risk of that envelope ripple yielding an audible and unpleasant burrr or trill to the end of notes that can sound exactly like distortion to some folks.
In a perfect world, there would be two isolated paths coming from the 3080, one for the rectifier and one for the audio, such that they didn't interact with each other.
The long and the short of it is that while JC's suggestion has a defensible basis, you may well be trading one source of distortion for another. Try it and see is all I can say.