Cannot escape an imaginary frequency

Questions
#1

In an optimization of acetone, I landed in an imaginary frequency (methyl rotations). I have been unable to escape it with a QChem input alone.

$comment
  Frequency job on previously-optimized structure. One imaginary frequency
$end

$molecule
  0       1
  O       0.0000166408    -1.3899581965     0.0000000000
  C      -0.0000007911    -0.1784689455     0.0000000000
  C      -1.2873334966     0.6118999407     0.0000000000
  C       1.2873091774     0.6119369775     0.0000000000
  H      -2.1389635781    -0.0719959657     0.0000000000
  H      -1.3366970362     1.2618953135    -0.8820470648
  H      -1.3366970362     1.2618953135     0.8820470648
  H       2.1389588978    -0.0719344727     0.0000000000
  H       1.3366540249     1.2619337616     0.8820470677
  H       1.3366540249     1.2619337616    -0.8820470677
$end

$rem
  JOBTYPE              FREQ
  METHOD               wB97M-V
  XC_GRID              3
  SCF_CONVERGENCE      10
  BASIS                pcseg-1
  THRESH               14
$end

@@@

$comment
  Optimization (reading in initial Hessian) +
  Analysis of optimized structure (to ensure we have escaped
  the imaginary frequency)..
$end

$molecule
  read
$end

$rem
  JOBTYPE              OPT
  METHOD               wB97M-V
  XC_GRID              3
  SCF_CONVERGENCE      10
  BASIS                pcseg-1
  THRESH               14
$end

$geom_opt
  INITIAL_HESSIAN            READ
  GRADIENT_CONVERGENCE       3e-5
  DISPLACEMENT_CONVERGENCE   1.2e-4
  ENERGY_CONVERGENCE         1e-7
  HESSIAN_VERIFY             RECOMPUTED
  FINAL_VIBRATIONAL_ANALYSIS TRUE
$end

I have tried tightening the SCF and optimization convergence thresholds, using a larger XC grid, doing a Freq + Opt + Freq job with the old OPTIMIZE driver, turning off symmetry, with no success. After a (large enough) manual displacement along the imaginary mode, I indeed managed to find a structure with no imaginary frequencies. I am curious why the optimizers are unable to do this, though.

#2

Sometimes when you land on a high-symmetry point (or even a point with local symmetry like the eclipsed conformation of a methyl rotation), the gradient may actually vanish. The solution is to write a script to displace along the normal mode with negative curvature.