If you look at the contents of the OpenJDK source packages, you’ll see that most of their source code is shipped inside tarballs inside the packages. If you run David A. Wheeler’s SLOCCount on the extracted package source only, it won’t count the code in those tarballs, so you’ll get a very partial count.
To properly compare the source code sizes, you’ll need to extract the tarballs before running sloccount. Doing this gives the following stats (on the versions I analysed):
openjdk-6 openjdk-7 openjdk-8
java: 2860304 (68.33%) 3232714 (72.32%) 3601973 (67.00%)
cpp: 648574 (15.49%) 671628 (15.03%) 892040 (16.59%)
ansic: 584813 (13.97%) 466742 (10.44%) 374378 (6.96%)
sh: 49935 (1.19%) 45608 (1.02%) 32891 (0.61%)
xml: 23937 (0.57%) 25927 (0.58%) 456037 (8.48%)
objc: 13062 (0.29%) 13557 (0.25%)
asm: 14125 (0.34%) 11555 (0.26%) 3508 (0.07%)
perl: 2258 (0.05%) 1220 (0.03%) 198 (0.00%)
pascal: 1089 (0.03%)
awk: 631 (0.02%) 687 (0.02%) 367 (0.01%)
python: 310 (0.01%) 310 (0.01%) 800 (0.01%)
haskell: 195 (0.00%) 195 (0.00%)
sed: 172 (0.00%)
cs: 72 (0.00%) 72 (0.00%) 72 (0.00%)
ruby: 44 (0.00%) 44 (0.00%)
jsp: 24 (0.00%) 24 (0.00%) 24 (0.00%)
csh: 3 (0.00%) 3 (0.00%) 3 (0.00%)
showing an increase in size (note however that the scope of the packages has changed with time too).