unit lost when divise (or multiply) same units

[bassjobsen]

test {
t1: (2em/1em) + 20;
t2: 20 + (2em/1em);
t3: 2em/1em;
t4: (2em/1px) + 20;
t5: 20 + (2em/1px);
t6: 2em/1px;
t7: (2em*1em) + 20;
t8: 20 + (2em*1em);
t9: 2em*1em;
t10: (2em*1px) + 20;
t11: 20 + (2em*1px);
t12: 2em*1px;
}

outputs:

test {
  t1: 22em;
  t2: 22;
  t3: 2em;
  t4: 22em;
  t5: 22em;
  t6: 2em;
  t7: 22em;
  t8: 22;
  t9: 2em;
  t10: 22em;
  t11: 22;
  t12: 2em;
}

The lost of units for t2 vs t5 and t8 vs t11 has already been described at #2471, but also t5 vs t11 seems unexpected.
Indeed mathematically em/em = 1 and em/px = ??, but with strict-units=off i expect both should be em.

[lukeapage]

in 1.7.5 the output is

test {
  t1: 22em;
  t2: 22;
  t3: 2em;
  t4: 22em;
  t5: 22em;
  t6: 2em;
  t7: 22em;
  t8: 22em;
  t9: 2em;
  t10: 22em;
  t11: 22em;
  t12: 2em;
}

in 1.3.0 before strict math

test {
  t1: 22em;
  t2: 22em;
  t3: 2em;
  t4: 22em;
  t5: 22em;
  t6: 2em;
  t7: 22em;
  t8: 22em;
  t9: 2em;
  t10: 22em;
  t11: 22em;
  t12: 2em;
}

[seven-phases-max]

+1.