Untested, but something like this
plot "<(sed -n -e 3p -e 7p -e 9p xy.dat)" using ...
Another option may be to annotate your datafile, if as it seems, it contains multiple datasets. Let's say you created your datafile like this:
1 2 3
2 1 3 # SetA
2 7 3 # SetB
2 2 1 # SetA SetB SetC
Then if you wanted just SetA you would use this sed command in the plot statement
sed -ne '/SetA/s/#.*//p' xy.dat
2 1 3
2 2 1
That says..."in general, don't print anything (-n), but, if you do see a line containing SetA, delete the hash sign and everything after it and print the line".
or if you wanted SetB, you would use
sed -ne '/SetB/s/#.*//p' xy.dat
2 7 3
2 2 1
or if you wanted the whole data file, but stripped of our comments
sed -e 's/#.*//' xy.dat
If you wanted SetB and SetC, use
sed -ne '/Set[BC]/s/#.*//p' xy.dat
2 7 3
2 2 1