There is not really a trade off required. Given the task which is to look for 'good' urls that match as closely as possible 1 or more words you want to do something like (pseurocode)

# get the domain part dropping www. and passing back the
# domain and tld (ie .com .net) or sld.tld (ie co.uk, com.au )
my ($domain, $tld) = get_domain( $url );

# chop domain into all possible substrings say 3-16 chars long, retrun
+ ary ref
# there are very few valid well known words > 16 chars, virtually none
+ > 24 chars
my $tokens = tokenize( $domain );

# get the possible words ordered by length(word) DESC ie longest first
# use a hash lookup or a RDBMS with a dynamicly generated SQL IN claus
+e
my $words = get_real_words_from_dict( $tokens )

# substitute out the words, as we remove longest first 
# we aviod finding substrings like 'be' in 'beetle'
my $residual = $domain;
my @words = ();
for my $word( @$words ) {
    # we may have duplicates of same word
    push @words, $word while $residual =~ s/\Q$word\E/;
}

# remove '-' from residual so 'come-here' will be two words, no residu
+al
$residual =~ s/-//g;

# work out % residual
$residual = 100*$residual/$domain;

# So now we have our data 
# @words 0..N is number of words found
# $residual is the %residual on that domain name
# $tld is the domain name

# say we inserted into a Db table like:
CREATE TABLE urls (
    url CHAR(75),
    words INT,
    residual FLOAT,
    tld CHAR(10),
)

"INSERT INTO urls (url,words, residual, tld) VALUES(?,?,?,?)",
    $url, scalar @words, $residual, $tld

You can now generate reports. Essentially you want something like:

SELECT url FROM urls
WHERE words >= 1
ORDER BY words ASC, residulal ASC
GROUP BY tld
[download]

This does not apply limits or add bias for say a pref for .com domain names. It will output urls with single word, lowest->highest residual first, then two words etc. Given what you want if the residual is > 10-20% you can probably just ignore those URLs and not insert them.