Do not emulate larger edges than we will actually use for this round of
MC. Decoding goes from avg+SE 29.972+/-0.023sec to 29.856+/-0.023, i.e.
0.12sec or ~0.4% faster.
(cherry picked from commit 44002d83)