Any test that does a moveChunk or split command is liable to run into issues because chunks are split unexpectedly.

For example:
1) bulk data insert
2) auto split is activated
3) manual moveChunk
4) assert on the count of data that moved fails because a smaller chunk was moved than the test expected.

We've seen ~5 BFs related to incorrect number of expected chunks, etc.
We could potentially have problems with > 50 sharding tests that didn't take auto splitting into account when written. We would also have to be very conscientious of auto splitting in future test writing if auto splitting were not disabled by default.