No, here is the relevant quote from the documentation for the strict pragma:

strict refs 

This generates a runtime error if you use symbolic references (see per
+lref).

...

There is one exception to this rule:
    $bar = \&{'foo'};
    &$bar;

is allowed so that goto &$AUTOLOAD would not break under stricture.
[download]

and here is code demonstrating the problem:

>perl -Mstrict -e "sub walk {} sub run {} my $walkable = defined(&{'wa
+lk'}); my $runnable = &{'run'};"
Can't use string ("run") as a subroutine ref while "strict refs" in us
+e at -e line 1.
[download]

So, the question is: How does the documented exception for use strict 'refs', which allows use of a reference to a symbolic sub, explain the observed fact that when defined is applied to the sub itself (not a reference), no runtime error is raised?

Or, conversely: If the expression &{'walk'} is legal as an argument to defined, why is the expression &{'run'} illegal as the right-hand-side of an assignment?

How (and why) does Perl distinguish between these two cases?

Ooooooh, well that could be called a bug

But if you ask me, both 'run' and 'walk' being string constants, well, I don't see why either &{'run'} or &{'walk'} should be illegal

Athanasius clarified what I wanted to say. Thanks a lot :) By the way, the following is equivalent to the code in question:

no strict 'refs';
my $walkable = defined *{ 'walk' }{CODE};
[download]

In this case, "no strict refs" is required and it seems reasonable. Should we cosider defined &{ 'walk' } not to be a recommended expression?