Split on the delimiter or regex into an array (the first arg of split is a regex). Then use that array as input to a grep in a scalar context which will give you a value of 0,1,2,3,4... If the value is 0, then there are no fields >255 and this is a "good line".

while (<IN>)
{
    chomp;               # needed if not splitting on whitespace

    my @fields = split;  # default split - splits on whitespace: /\s+/
                         # otherwise use: 
                         #   my @fields = split(/regex_here/, $_);

    if ( grep{length($_) >255}@fields )
    {
        print "BAD LINE: $_\n"; #do what you need here
    }
    else
    { 
        #...its a "good line" here
    }
}
[download]

updated: minor formatting tweak

Given that with the "sqlldr" method you would have to delete the "bad" data after it was discovered to be bad, it seems like you'd be wasting all that effort. Using Perl to validate the data before you load it from the flat file would be trivially easy and would avoid the wasted "round trips".

# Checks comma-separated elements
perl -F, -wane'for (@F){die "Improper data\n" if length($_)>255}' file
+name.txt
[download]

Assuming that your delimiter is in $delimiter then:

if ( $row =~ /(?:^|\Q$delimiter\E)[^\Q$delimiter\E]{256,}(?:\Q$delimit
+er\E|$)/ ) {
    print "Field greater than 255 characters in Row:\n\t$row\n";
    }
[download]

I guess Oracle's import and checking is faster than if you do it in Perl. But in the end only comparison will show if that's true.