The rationale provided is:

1. Promoting two different ways to exchange the same data forces Clients to support both structures and complicates future FHIR Write efforts.
2. Transforming QuestionnaireResponse to Observations is necessary if a system only supports Observations.
3. Clients can not query individual responses in QuestionnaireResponse.
4. Potential clinical safety issues with information spread across two resources.

I believe these are all flawed.

Questionnaires is the only acceptable way to capture a significant number of assessments.  Only Questionnaires provide strict control over question wording, permitted answers, order of presentation, flow logic, answer weighting and scoring, etc.  It is not possible to achieve these same things using Observations.  Many assessments are very carefully designed and validated instruments.  Failing to adhere to the strict requirements about presentation means that the data collected is suspect and may be invalid.

Data has long been captured in existing systems as both completed forms as well as a subset of relevant data (typically scores) then represented as computably retrievable observations.  FHIR offers the ability to move from scanned paper to electronic entry, but the utility of retaining both the original "source of truth" completed form as well as the smaller set of extracted Observation data remains.

There has likely been some degree of patient safety issues in the past with transcribing scores into electronic systems.  However, with electronic QuestionnaireResponse representation and data extraction mechanisms, this downside is significantly reduced.  And, given the risk certainly wasn't enough to drive systems to stop using proper forms before, it certainly doesn't make sense to move away from them now that the risk is lower.

In the old world, the source-of-truth data was never searchable based on answers, so the inability to do so in the new world isn't problematic either.  There's generally no need to capture every single answer as an Observation.  Only certain key questions or the final determination need to be computably accessible.  

Numerous IGs that build on U.S. core require support for Questionnaire and QuestionnaireResponse, including Gravity and Da Vinci DTR and CDex.  Much broader consultation is needed before making such a major shift.