I've never seen a colliding pair in a "true" hash function (not just CRC) be of differing lengths. They've always been a pair of blocks with very few differing bits, such that the differences "cancel out" in the computation and the internal state is the same regardless of which of the pair is run through it. I suspect trying to create a colliding pair with differing lengths is much more difficult, since SHA1 also includes the length as part of the calculation.
