Actually, the list of values from the RHS is also a list of aliases (to values or to scalar variables, depending on what semantics you want to use at this point).

[The list for the LHS is a list of aliases to variables (not values) but that only makes a difference for aggregate variables (arrays and hashes) -- for assigning from a list of scalar variables to a list of scalar variables, the two lists that get constructed contain identical types of aliases.]

At the time that the code is compiled, Perl tries to determine if it is impossible for the same alias to end up in both lists. If Perl can't determine that that is impossible, then the AAssign op is flagged as "COMMON" (at compile time).

When an AAssign op is executed, if it is flagged as "COMMON", then, before it does the assignments, it first replaces each item in the RHS list (of aliases) with a (non-alias) copy of itself.

 $ diff <( perl -MO=Concise -e'($a,$b)=($b,$a)' \
                                      | perl -pe's/0x[\da-f]+/0x/g' ) 
+\
        <( perl -MO=Concise -e'($x,$y)=($b,$a)' \
                                      | perl -pe's/0x[\da-f]+/0x/g' )
-e syntax OK
-e syntax OK
4c4
< 9     <2> aassign[t5] vKS/COMMON ->a
---
> 9     <2> aassign[t5] vKS ->a
14c14
< 7              <#> gvsv[*a] s ->8
---
> 7              <#> gvsv[*x] s ->8
16c16
< 8              <#> gvsv[*b] s ->9
---
> 8              <#> gvsv[*y] s ->9
[download]

That is a great explanation! Zoomed in :)