generally lower the e-value more significant is the hit ... database size generally determines the evalue cut-off ... further if you are expecting your sequence to be a lot more diverse than the sequences in database then evalue will be higher ... have a look at this link https://www.biostars.org/p/17326/ and at this paper http://bioinformatics.oxfordjournals.org/content/25/6/824.full for more information

you can always filter the top hit which is generally the best hit from your results and leave out the one's with lot of mismatches and gaps as less confidence hit ... these will also be reflected in the high e-value

There is the old rule-of-thumb that an e-value less than 0.1 means that the hit is significant. This was however true some 15 years ago but it does not hold anymore when you search against a real big databank like the NCBI nr.

A remark : when you search a small oligonucleotide against a complete genome or transcriptome in order to test the specificity of a primer or siRNA it is recommended to set the cutoff high (100 or even 1000, de facto not using a cutoff). The reason is that here you really want to find all sequences that give a 100% or close to 100% hit, since all are really significant, the probability of finding a hit when searching a random databank of same size is not relevant.

It is depend on your sample. If your sample is  fully sequenced and available in the database you can choose 1e -4 to 1e-10 (lower more confidence). If it is not in the database,  you can choose 1e-30. In this case you increase the number of hit and then you can select  the best hit based on what you are seeing.

Sometimes the e value is very low, but there is still mismatches. Please be aware of that.