I think a good way to do this is to write a Perl script selecting the files larger than, say, 20M; and do not care with the sorting in it. Then you can easily sort|head the resulting few files.

By the way, I have some old code doing something similar. I have a file with a lot of records, and I have to select the 64 "best" records. Here's the code, with omissions (##)

## ... HEADER STUFF ...

@sellpt= (0)x64;
@sellx= @sellpt;
@selly= @sellpt;

## ... MORE HEADER STUFF, OPENING HANDLE "VAR" ...

while (<VAR>) {

        my($p,$x,$y);

        ## FILLING ($x,$y,$p) WITH DATA FROM RECORDS, MUST GET THE ONE
+S WITH LARGEST $p

        $sellpt[-1]<$p and do {
                my ($u, $v)= (0, 0+@sellpt);
                defined ($sellpt[int(($u+$v)/2)]) or
                        die qq([D30] $u $v (@sellpt));
                while ($u<$v) {
                        if ($p<$sellpt[int(($u+$v)/2)])
                                { $u= int(($u+$v)/2)+1 }
                        else
                                { $v= int(($u+$v)/2) };
                        }
                @sellpt= (@sellpt[0..$u-1], $p, @sellpt[$u..@sellpt-2]
+);
                @sellx= (@sellx[0..$u-1], $x, @sellx[$u..@sellx-2]);
                @selly= (@selly[0..$u-1], $y, @selly[$u..@selly-2]);
                };

        ## ... SOME MORE STUFF ...

        };

## ... AND SOME MORE, WRITING OUT @sellx,@selly IN SUITABLE FORMAT ...
[download]