rebuild-index --rebuildAll corrupts the indexes for certain data sets

For certain data sets, running rebuild-index --rebuildAll causes entries to be removed from the index. The entries are still present in the backend, but not available through the index.

Reproduction steps:

1) Start OpenDJ
2) Rebuild indexes
3) Add a lot of entries using LDAP.
4) Verify indexes (The indexes are still good at this point.)
5) Run rebuild-index --rebuildAll
6) Check the error log. See that when the backend is brought online after the index rebuild, the number of entries is less than when the rebuild started.
7) Verify indexes. See that many errors are reported.

Notes:
I could not reproduce this problem using the setup --sampleData option to create the data, only adding the data through LDAP.
When I skipped the initial rebuild indexes before the data was imported, I could not reproduce the problem.
I have only reproduced this problem with very large data sets, 150,000+ users.
Running rebuild-index again after the corruption seems to fix the index, and subsequent rebuild-index calls do not break it again.
The errors returned by verify indexes seem to indicate that sequential blocks of users were lost.