s/([^[:^alpha:]nN])([\dnN]+\1)/_$1$2_/g;
s/__/_/g;
[download]

Update: Ah, you had more specs hidden in your code block (and roughly that solution, except without the double-^ trick that I like). You can't use things like \3 in a character class, even if you don't try to use them out-of-order. I'll take a stab later. Update:

my $string = '423RY75Y69827EC67592C78657N965R345U299M';
my $correct= '423R_Y75Y_69827E_C67592C_78657N965R_345U_299M';
for( $string ) {
    s/([^[:^alpha:]nN])([\dnN]+\1)/_$1$2_/g;
    s/(?<=([^\dnN_]))?([\dnN]+)([^\dnN_])/
        $2 . $3 . ( $1 eq $3 ? '' : '_' )
    /ge;
    s/_(_|$)/$1/g;
}
print "Okay\n"   if  $string eq $correct;
[download]

Add "defined $1 &&" or "no warnings 'uninitialized'" to avoid a warning, if you need to.