One perl solution if there is no overlap:

$Ln=1000;
for ($i=int(rand(50))+1;$i<=$Ln;$i=$i+50) {
        printf "%d\n",$i+1-int(rand(3));
}

I'd follow this approach:

1. Allowable length = 48 .. 52
2. Iterate through each sequence. For each sequence,
   • if len(seq)<52, skip to next sequence. Else,
   • pick a random number (call it point) between 0 and len(seq)-1
   • if len(seq)-1 - point >=52, pick substring 3' of point, with length randomly picked between 48 and 52
   • add to a new list "pool_1" the sequence 5' of point and 3' of point+length picked above (these are the flanking fragments)
3. Repeat above operation on pool_1, this time picking substrings 5' of chosen point and adding the fragments to "pool_2".
4. Repeat until both pool_1 and pool_2 are filled with fragments less than 52 in total length

If you're concerned with performance, use unpack instead of substring.