In my opinion, if the designer or the GM doesn't like the overall probability curve of one (or more) of the clocks, the answer is to iterate and redesign the curve, not curate out particular content produced by the content introduction machinery (which the players are interfacing with to make their general and system-relevant guile based decisions).
Everything else being equal that's good advice. But it's also hard! And if you use randomly-timed content introduction (ie wandering monsters) then there is always a chance that you will get improbably severe results even when the players haven't deserved the punishment thus inflicted.
In discussions of Cortex+ play (including in the Hacker's Guide) there is the suggestion that the GM not always build the best pool s/he can out of a Doom Pool roll if doing something else would better serve the trajectory of the game. Frankly that's a concession to weaknesses in the design of the Doom Pool as a system (it serves many different functions and it's hard to make all of them work all of the time) but maybe the designers just couldn't get it any better and still keep it workable.
I see that advice as somewhat similar to Gygax's advice about the wandering monsters - if the players don't deserve more punishment then don't roll (or ignore the result - same diff) even if the rules say you should. In neither case is it force (in my view), because in neither case is it adjusting or manipulating the outcomes of action resolution. (This is why I see Gygax's insistence on
not letting the PCs escape unnaturally as key - because that would be subverting action resolution.)
As far as his comments about allowing the PC to be maimed rather than killed, as I said that's barely force because it's barely manipulation. To make it not be force at all, all you need is a rule (both 4e D&D and Prinve Valiant have versions of this) that says "If you want, zero hp can be some sort of incapacitation shot of death"). That's a pretty trivial change. And Gygax is very clear that this sort of thing should not be done so as to fundamentally alter what was at stake in play ("disinterest", "always give the monster an even break"). It's only removing death as the only failure state - but I don't think that's fundamental to classic skilled play D&D, as he says (pointing to the existence of resurrection magic).
Whether any of these things is good or bad GMing is a different matter, but I think the nuance with which Gygax addresses them is one of his high points in grasping what is going on with his game design and where it does or doesn't have capacity to give a little bit. It's much more subtle than I sometimes see suggested when people just present the quote about it being a GM's prerogative to change or ignore a die result. And it's more subtle than the AD&D 2nd ed passage that was presented upthread.
In my opinion, this is why both Moldvay Basic and Torchbearer are both just fundamentally better game's than Gygax's D&D (when it comes to challenge-based gaming).
I don't think Moldvay is fundamentally different on this particular issue. There is a remark somewhere in there about fudging, I think, though I can't recall the details.
Maiming rather than death is largely irrelevant in Moldvay because at levels 1 to 3 there is no regeneration magic and hence no recovery from maiming any more than from death. As far as the wandering monster issue is concerned, exactly the same thing can happen - ie there is a chance that a very well played party might nevertheless be absolutely hammered by wandering monsters while heading through the dungeon to their exploration goal, putting their punishment severely out of whack with what they deserve. Moldvay has no better way to correct this problem than Gygax does.