Could you provide a more information? Are your reads all from one sample and what format are they in (fastq, fasta)? Conservation of 6/30 nucleotides seems pretty low.

They are different genes in fasta format. I could also use a shorter window, about 20 nt.

Do you have 3000 sequences of 30 nucleotides from 3000 different samples? If yes you can do apply any multiple alignment tools, such as CustalW or another ones. If you guess that only 6 nucleotides are conserved to align such sequences will be quite difficult. In case that they are different genes I have doubt that it will be correct to align such sequences together.

U can use muscle or clustalw.

For generic alignment I have had good experiences with tcoffee (http://tcoffee.crg.cat/apps/tcoffee/do:regular), but I don't know if it is well suited for short sequences like yours. Would bowtie (http://bowtie-bio.sourceforge.net/index.shtml) work for this?

T-Coffe doesn't work with large sequence datasets. The best way is to use fast, accurate and iterative methods like MUSCLE or PROBCONS. MSAProb says having more accuracy than Muscle and Probcons but I did not try it yet! When run Probcons include iterative refinement option -ir 1000 (default 100) that is the maximum number of iterations. In uscle you also have this option by calling -maxiters (i.e 100, default 16). Good luck!