Unless I'm missing something, ... (as long as it's not necessary to match overlapping matches):

You hit the nail on the head. You'll only match 5,015,229 times when the OPs code matches 35,106,546 times.

However, with a modification to your regex, you can avoid that problem and find overlapping matches:

++$index{ $1 } while $$rSeq =~ m[(?=([ACGT]{7}))]g;
[download]

But it is still much slower than avoiding the regex engine completely:

[ 6:55:26.00] C:\test\humanGenome>..\976237 chr21.fa
16384
Using custom indexing found 35106546 matches; took 31.611258 seconds
Using custom index2 found 35106546 matches; took 27.504099 seconds
Using custom index3 found 35106546 matches; took 27.571143 seconds
Using quantified charclass lookahead found 35106546 matches; took 49.9
+54810 seconds
[download]

But ++ for thinking outside the box. (I can't believe I actually used that phrase :)