For short matches such as these you will need to increase the E-value cut-off, and possibly make the search more stringent by increasing the gap costs and the match/mismatch scores.

Note that switching to an alternative algorithm may also help in this case, I suggest looking at the FASTA suite programs (see here), in particular GLSEARCH and FASTM may be of interest.

Alternatively if you are not interested in statistics, but only want to know if a match does or does not occur, pattern matching methods such as those provided by the EMBOSS programs dreg and fuzznuc may be more appropriate.