> + final int length = (int) blob.length();

This call is additional roudtrip to the server, which quite often is more
expensive than allocating byte arrays in VM. Anyway, server send data in
chunks, max. of 64k size (and it is controlled by the blobBufferSize
property), so reading it into one big array instead of reading it in small
chunks should not be faster since it will be the same amount of roundtrips
to the server.

Did you see increase in performance?

Roman