Yes, it's a softfork for both sides, not a hard one. One side will eventually win due to variance.
It is barely even a soft fork.
For 2015 out of every 2016 blocks, the longest chain is effectively defined as the one with the highest block height.
The fork would need cross a difficulty boundary update. Even then, both sides would probably agree which of the 2 forks has the higher weight block. The only exception is if the rounding error causes the finite precision maths to consider both forks to be equal.
Consider two forks that happen exactly at the diff update point. For one fork (A), the difference in timestamps is 1209600 seconds (so 2 weeks). On the other fork (B), the difference is 1209599 seconds, so one second less.
This means that for A, there is no diff update at all. Both rules would agree here. For B, the difficulty rises by 1209600 / 1209599. Both sides will agree that fork B has higher difficulty, since difficulty is calculated to 256 bits of accuracy.
They will both break any tie in favour of fork B.