Why does 'grep -q' consume the whole input file?

grep does stop early, but it buffers its input so your test is too short (and yes, I realise my test is imperfect since it's not seekable):

seq 1 10000 | (grep -q 2; cat)

starts at 6776 on my system. That matches the 32KiB buffer used by default in GNU grep:

seq 1 6775 | wc

outputs

   6775    6775   32768

Note that POSIX only mentions performance improvements

When searching several files

That doesn't set any expectations up for performance improvements due to partially reading a single file.

This is obviously due to buffering that grep does to speed up things. There are tools which are specifically designed to read as many characters as requested and no more. One of them is expect:

{ expect -c "log_user 0; expect 2"; cat; } < infile

I don't have a system to try this on, but I believe expect will eat up everything until it encounters the expected string (2), and then terminate, leaving the rest of the input for cat.