>> but then later you posted a different patch.  Anyway FWIW here is a 
>> safer workable workaround that implements your first suggestion and 
>> sets a new explicit buffer-local variable in completing minibuffers, so 
>> modes that need to distinguish such minibuffers could check it.
>
> I combined your idea with that of Gregory for the patch below. It's far 
> from perfect, but I think it strikes a good balance of simplicity, 
> preserving compatibility, and moving in the right direction.
>
> WDYT?
>

I really like the simplicity of that solution, but (you know me, there's 
always a but...) I do not see what the next step in that direction could 
be.  My fear is that package authors will be encouraged to use a similar 
duct tape in their code, and that making the behavior of 
read-from-minibuffer depend on a environment variable (or an additional 
parameter) will never happen.

I attach a patch which is, I believe, safe in the sense that it cannot 
break any package, and which creates a transition period after which 
minibuffer-completion-table will automatically become buffer-local upon 
entering the minibuffer.