Blastx : huge difference between qcovs and (len/qlen)*3

Question

[Metaxo]

I was using blasts and trying to filter based on query coverage. I tried to use the built-in qcovs parameter, and also used (3len/qlen) at the csv output to confirm the results. For around 95% of the results, the two values are fairly similar to one another. However, for the remaining 5, qcovs is very high (70 or more) but the (3len/qlen) value is extremely low (around 10%) and I wonder why this is happening. Why can qcovs still be that high when the hit length is significantly lower than the query length?

Many thanks. Han